beads/CHANGELOG.md
Julian Knutsen f3c47390e7
fix(storage): refuse graph_mode workspaces on library opens (#7721)
#7671 fences graph workspaces in cmd/bd: command admission, the store
factories and the other CLI paths refuse one, and #7722 extends that
fence to a .beads holding the graph-preview-format marker whatever its
metadata says. But the public library opens (beads.OpenBestAvailable,
OpenBestAvailableWith, OpenFromConfig, OpenGated) and the Dolt store's
metadata-driven constructors (NewFromConfig, NewFromConfigWithOptions,
NewFromConfigWithCLIOptions) still opened either kind of graph workspace
as an Issue store and ran Issue schema migrations into the graph
database (or, with no metadata beside the marker, created an Issue
database inside the graph workspace).

- Move the typed refusal into internal/configfile
  (UnsupportedWorkspaceFormatError with BeadsDir, GraphMode and
  FormatMarker; CheckWorkspaceFormat; GraphFormatMarkerFile;
  HasGraphFormatMarker) with byte-identical text and hints for both the
  graph_mode and the marker refusal, and move cmd/bd's discovery
  pre-checks there as LoadAndCheckWorkspaceFormat and
  CheckDiscoveredWorkspaceFormat (LoadForDiscovery, which never migrates
  a legacy config.json; unreadable metadata beside the marker is refused
  as the marker's format). cmd/bd's checkWorkspaceFormat,
  loadAndCheckWorkspaceFormat, hasGraphFormatMarker and
  checkDiscoveredWorkspaceFormat become thin delegates, so the CLI keeps
  its behaviour (exit 16 with the same hints) and every caller of them is
  unchanged; bootstrap's ancestor walk builds the configfile type.
- Every public metadata-driven open runs CheckDiscoveredWorkspaceFormat
  as its first statement: before OpenGated takes its gates and before
  OpenBestAvailableWith's Load can migrate config.json. A refused open
  leaves the workspace byte-identical. The type is exported as
  beads.UnsupportedWorkspaceFormatError.
- The Dolt constructors run CheckDiscoveredWorkspaceFormat before
  configfile.Load, so a graph workspace marked only by a legacy
  config.json, or by the marker beside stripped, missing or unparseable
  metadata, is refused before Load can migrate it into metadata.json
  (and drop graph_* keys this build does not know). requireDoltBackend
  re-checks the loaded config as a backstop, so they also refuse before
  resolving a server or writing a port file or Dolt root.

Tests: configfile unit tests for both refusals (exact marker text and
hint, graph_mode winning over the marker, unparseable metadata with and
without the marker); the public opens and the three Dolt constructors
against graph_mode workspaces and against marker-only, stripped
(metadata.json and legacy config.json), server, explicit dependency and
unparseable-metadata marker workspaces, each refused with the typed
error and the workspace left unchanged.

Agent-Signature: claude-code-claude-opus-5-5-unknown-reasoning on behalf of Julian Knutsen

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-20 09:29:54 +00:00

731 KiB
Raw Permalink Blame History

Changelog

All notable changes to the beads project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Added

  • bd versions <id> lists the versions recorded for a bead, and versioned-history.enabled turns recording on. While it is on, every accepted change to a bead created while it is on, or that enters the issues plane while it is on (a wisp promoted or made persistent), stores a snapshot of it as a row of issue_versions in the same transaction. It is off by default and nothing is backfilled: a bead already on the issues plane when it is turned on is legacy, and editing it records nothing. A recorded write leaves updated_at where an unrecorded one would, so bd import's stale check sees the same timestamps either way. bd versions --json answers {"versions": [...], "recording": bool}; the command still lists what was recorded after recording is turned off, and proxied-server and HTTP workspaces refuse it. The setting is a row of each store's config table (bd config set versioned-history.enabled true), and every store reads its own when it is opened: a write routed into another workspace (a prefix-routed bd update or bd close, bd create --repo) follows the target store's row, a proxied-server provider reads it when it is constructed, and a long-lived bd serve keeps the value it read until it restarts. BD_VERSIONED_HISTORY_ENABLED=1 (or a config.yaml value) turns recording on for every store the process opens, and never turns it off. The row replicates on bd dolt push/pull, and recording is single-writer only: two clones recording at once can mint the same revision and fail the pull, and a write from a writer that is not recording is folded into the next recorded version under that version's actor, so keep recording on for every writer of a store or off for all of them. Turning it on with bd config set first checks every issue the store would version and refuses, writing nothing, while any holds a number outside the I-JSON exact-integer range (magnitude above 2^53-1, however it is spelled) in its metadata or in a gate's timeout (held as nanoseconds, so about 104.25 days), and prints the count, the ids, the field and the fix. There is no override, and the check runs once; after that, a recorded write that would hold such a number fails atomically. Migration 0071 adds a nullable removed_restriction column to issue_versions. bd version now refuses an argument and suggests bd versions <arg>, instead of printing bd's build for a mistyped bd version bd-123.

  • bd blocked --flags prints the blocked flag for each id read from stdin (one per line; blank lines and repeats ignored), with --json as [{"id","is_blocked"}] in the order the ids were given, omitting an id that names no row on either plane. A closed id is answered too, with its stored flag (a status <> 'closed' SQL read of the column leaves it out). It is the flag bd ready filters on, not the listing: a row whose stored flag is set with no nameable blocker reads true here and is absent from bd blocked. It runs issueops.BlockedFlags on every route; over HTTP that is listBlockingAnnotations with is_blocked, behind dependencies.blocking.isBlocked, refused before the dial against a server that does not advertise it. The listing's filter flags are refused with --flags.

  • bd blocked --flags --raw answers the stored is_blocked column itself, without the external-dependency policy: a row blocked only by an unsatisfied external: ref reads false (the policy-aware --flags reads true); a stale flag reads true either way. --raw is for the local and proxied routes, where it reads the column directly and the two answer the same. Over HTTP it is answered only by a server that does not advertise policy.external_dependencies (whose flag is the column); a server advertising it answers only the policy-aware flag, so --raw against it is refused before the dial rather than answered with the decorated flag. --raw without --flags is refused.

  • bd blocked works over a connected HTTP workspace. The blocked listing is a role, issueops.BlockedLister (ListBlocked(BlockedRequest{ParentID, Labels, LabelsAny, ExcludeLabels}) -> BlockedResult{Items}), on every backend leg over the same body GetBlockedIssues runs, and bd serve publishes it as GET /v0/beads/blocked (parent, and the repeatable label, label_any and exclude_label), answering {items, has_more} with BlockedIssue rows (the element bd blocked --json prints), behind the new blocked.list capability token (additive; no wire_revision bump). The listing names unsatisfied external: blockers wherever the external-dependency policy applies: on the server when it advertises policy.external_dependencies, else on the client, as every other policy read does. The HTTP backend's BlockedLister() and legacy GetBlockedIssues dial it; against a server that does not advertise the token they refuse before dialing with a typed capability error naming blocked.list, and never compose the listing from other reads. The listing is unpaged: a server answer that reports has_more is refused with the new httpclient.TruncatedListingError rather than returned short, and an answer over the client's 32 MiB response limit is refused as any oversized response is. The external-dependency decorator's legacy GetBlockedIssues over an HTTP store now reads the extra sources through the served roles (it failed on a raw multi-id read before), so callers of that door no longer fail on it; commands that also make other raw reads the HTTP backend does not serve (bd ready --explain) still refuse on those. bd blocked runs the role on every route (the proxied-server route through the provider's accessor) with unchanged output: --json prints the same rows in the same order (priority, then newest first, then id, which now also breaks a tie the order used to leave to chance), and [], never null, when nothing is blocked.

  • bd list's remaining filters work over a connected HTTP workspace. GET /v0/beads/issues accepts updated_after/updated_before, closed_after/closed_before, defer_after/defer_before, due_after/due_before, priority, priority_min, priority_max, title_contains, desc_contains, notes_contains, external_contains, external_ref, spec, label_pattern, label_regex, empty_description, no_labels, no_parent, a repeatable exclude_type, pinned, no_pinned, deferred, overdue, mol_type and wisp_type, behind one new capability token, issues.list.filters (additive; no wire_revision bump; the names are issues:count's where it already spells the filter), and applies them through the same Reader role a local bd list reads through. Bounds are strict and compared at whole-second precision on both routes; relative and date-only values are resolved on the client's clock and timezone as locally, while --overdue compares against the server's clock. Against a server that does not advertise the token the client refuses before dialing, naming issues.list.filters, rather than sending a filter an older server rejects or dropping one and answering a wider listing. The 29 corresponding divergence-ledger refusal rows are removed. The default bd list --parent tree view does not carry these filters yet: its descendant walk still refuses one over HTTP rather than drop it, and --flat or --json skips the walk and applies them to the parent's direct children (not its whole subtree).

  • issueops.ListRequest.ExcludeStatus: statuses to leave out of a listing, appended to the default exclusions (never re-admitting what they hide), comma-separated or repeated, each validated against the workspace's statuses like CountRequest.ExcludeStatus (an unknown name is ErrValidation). GET /v0/beads/issues accepts it as a repeatable exclude_status under the same issues.list.filters token (an unknown name is a 400 on exclude_status), so an HTTP-backed caller can leave closed history out of an all listing (all plus exclude_status=closed keeps every status not named, the pinned and hooked statuses and custom done/frozen statuses included) instead of reading the whole history. It is refused together with ReadyFlag, like the other members the ready query cannot carry. There is no bd list flag for it yet.

  • bd serve publishes the gate members: await_type, await_id, timeout (nanoseconds) and waiters on POST /v0/beads/issues and on issues:batchApply create items, behind the new issues.create.gates capability token, and await_id and waiters (a whole-value replacement; [] clears) on the patch of PATCH /v0/beads/issues/{id} and of issues:batchApply update items, behind issues.update.gates. Additive; no wire_revision bump. The bounds are the library role's, the same ones a local create or update now enforces (see Changed), answered as 400 invalid_argument with param naming the member.

  • issueops.IssuePatch.Waiters: a whole-value replacement of a gate's waiter list (nil or empty clears it, stored as []), on Lifecycle.Update and on BatchApplier update items, on every backend. issueops.GateFieldError (Field, Detail; wraps ErrValidation) names the gate member a create or update refused, as one of the new issueops.GateFieldAwaitType, GateFieldAwaitID, GateFieldTimeout and GateFieldWaiters constants.

  • The HTTP client sends the gate members on create, update and issues:batchApply (create and update items) to a server that advertises issues.create.gates / issues.update.gates. Against an older server a request naming any gate member is refused before the dial with a typed capability error naming the missing token (never sent without the member); a batchApply plan is checked whole, so no partial plan is sent. A request naming no gate member is unaffected by either token.

  • bd heartbeat works over a connected HTTP workspace. The lease refresh is a role, issueops.Heartbeater (Heartbeat(HeartbeatRequest{Actor, IssueID}) -> HeartbeatResult{IssueID, Holder, HeartbeatAt, LeaseExpiresAt, Rearmed}), on every backend leg, and bd serve publishes it as POST /v0/beads/issues/{id}:heartbeat behind the new issues.heartbeat capability token (additive; no wire_revision bump), with the new lease_not_held (409) problem code (reason: ephemeral for a wisp). A refresh writes ONLY the lease row: the issue's revision is not reminted (a guarded write against the pre-heartbeat token still lands), no version, event, journal entry or Dolt commit is recorded, and no hook fires. Rearmed is true only when the heartbeat created the caller's lease row (a hand-doled claim's first heartbeat). A re-arm is recorded as a mutation; a refresh is not: the re-arm remints revision, stamps updated_at, records a lease_rearmed event, a journal entry and a version, makes a Dolt commit and fires on_update (see Fixed); the reported times are whole seconds, exactly what the lease row stores. The HTTP backend's Heartbeater() and legacy HeartbeatIssue dial it; against a server that does not advertise the token they refuse before dialing with a typed capability error naming issues.heartbeat, and never fall back to a re-claim or a metadata stamp. A busy (503) answer is replayed, because a repeat only moves the expiry later. bd heartbeat now runs the role on every route (the proxied-server route through the provider's accessor), with the same {"id","status":"heartbeat","owner"} success output and the same exit codes. Refusal TEXT on stderr changes per route: the direct routes (Dolt server, embedded) gain a lease not held: prefix (heartbeat X: lease not held: issue not claimable: X status closed); the proxied-server route loses its Heartbeat: db: IssueSQLRepository.HeartbeatIssue: prefix and reads as the direct routes do; over HTTP the refusal is the problem's detail, which names the state found (it is closed, an ephemeral wisp, open and not claimed) or, for a foreign holder, the native issue already claimed by <holder>. An id that names no row answers not-found from the role rather than "not claimable" (the CLI resolves the id first, so its output is unchanged), and every refusal for a caller that holds no live claim (closed, released, reclaimed, moved off in_progress, or a wisp) is issueops.ErrLeaseNotHeld, which still matches ErrNotClaimable.

  • GET /v0/beads/dependencies/blocking accepts is_blocked, behind the new dependencies.blocking.isBlocked capability token (additive; no wire_revision bump), and answers each item's stored, transitive is_blocked flag, the column bd ready filters on; an item whose id names no row carries none. issueops.BlockingRequest gains IncludeIsBlocked on every BlockingAnnotator implementation, and issueops.BlockedFlags is the batched read over it. The HTTP backend's IsBlockedBatch now serves through it; against a server that does not advertise the token it still refuses before dialing, now with a typed capability error naming the token. An unsatisfied external: blocker sets the flag on every route, just as it keeps the issue out of bd ready.

  • bd promote works over a connected HTTP workspace (S21b). Its one library entry, issueops.Promote(ctx, PromoteRoles{Reader, Lifecycle, Commenter}, PromoteRequest), reads the row, refuses one that is not a wisp (*issueops.NotAWispError, the sentence bd promote always printed), moves it with ONE Lifecycle.Update whose patch is Persistence = persistent (labels, dependencies, events, comments and the child counter move with it in the role's transaction), then records the promotion comment. The direct and proxied-server routes run it over their own roles, so a local promote now makes the same plane move bd update --persistent makes, honoring dolt.auto-commit. That changes five things a script can see on those routes: the workspace's on_update hook fires twice, once for the move and once for the comment (the direct route never fired it, the proxied-server route fired it once); the promotion comment is a row bd comments lists (the direct route recorded only a commented event); a promoted parent keeps its child counter, so its next bd create --parent mints the number after its wisp children's instead of failing with issue already exists: <parent>.1; a promotion records two Dolt commits, bd: promote <id> for the move and bd: comment <id> for the comment (an embedded workspace recorded one, bd: promote (auto-commit), and the proxied-server route one, bd: promote <id>); and a promote that loses a race to a concurrent one fails with the not-a-wisp sentence instead of promoting <id>: wisp <id> not found. A comment that cannot be written leaves the promotion standing behind a warning on every route; the proxied-server route used to roll the promotion back. bd serve publishes the move as patch.persistence on PATCH /v0/beads/issues/{id} behind the new issues.update.persistence capability (additive; wire_revision is unchanged), which also carries bd update --persistent, --ephemeral and --no-history over HTTP. Against a server without the token the client refuses before the dial, naming it; nothing is written. A move the workspace refuses (an unversioned-class row asked for --ephemeral or --no-history, or a demotion of a row with retained snapshots) matches ErrValidation on every route: bd serve answers it 400 invalid_argument, and the local refusal gains a validation failed: prefix.

  • bd serve advertises the new issues.update.claim behavior token for what PATCH /v0/beads/issues/{id} already served untokened: claim (upstream #6890), resolved on both planes, so a wisp is claimable there (additive; no wire_revision bump). The HTTP client now picks the claim route from the handshake before dialing: with the token every claim is one updateIssue call; without it a claim alone goes to {id}:claim (no doomed 400 first; a wisp still refuses by name, W-ClaimRequest.Wisp, now naming the token), and a claim beside a patch, a guard, a force override or allow_template is refused before the dial with an unsupported error naming issues.update.claim. Skew note: a server that serves claim untokened (wire_revision 2 builds before this release) now gets those two refusals from a new client where it used to serve them; upgrade bd serve. A template claimed alone without the token is still refused, on the client's own pre-read: {id}:claim has no template guard (bd-jkp9v3).

  • issueops.LeaseReclaimer, the stale-lease sweep behind bd reclaim, is a role on every leg (dolt, embedded dolt, the unit-of-work provider and the HTTP client), and bd serve publishes it as POST /v0/beads/issues:reclaim behind the new issues.reclaim capability (additive; wire_revision is unchanged). Every reverted row mints a new revision, reported per entry as revision, so a holder that writes with its pre-reclaim token gets a version mismatch. An id that is not a stale lease is left out of the answer, not refused. bd reclaim now runs the role on both routes, which changes four things a script can see: its --json entries gain revision; the workspace's on_update hook fires once per reverted row (the raw path fired none); --id now accepts at most 1000 ids per run (more is refused, never truncated); and the role records its own version commit, honoring dolt.auto-commit the way bd prune does, as bd: reclaim N expired lease(s) on both routes. Over a connected HTTP workspace bd reclaim now works instead of refusing.

  • bd show --comments-tail N renders only the last N comments in text output (including under --watch) (#6618), preceded by one elision line naming how many older ones were hidden. A render-only cap for fat, append-only beads whose full comment history is hundreds of KB — description and metadata are unchanged, and omitting the flag (or passing 0) is byte-identical to today's output. JSON output is untouched; --include-comments still streams every comment there.

  • GET /v0/beads/issues accepts no_assignee, title and a repeatable id (one id per value, at most 200), each behind its own new capability token (issues.list.noAssignee, issues.list.title, issues.list.ids; additive, no wire_revision bump; id and its token are bd-enterprise's contract, adopted unchanged), and applies them through the same Reader role a local bd list reads through. Over a connected HTTP workspace bd list --no-assignee, --title and --id now work instead of refusing. Against a server that does not advertise a flag's token the client refuses before dialing, and the refusal names the token (every capability refusal over HTTP now names its token).

  • GET /v0/beads/issues accepts q, bd search's free text, behind the new issues.list.search capability token (additive; no wire_revision bump). issueops.ListRequest gains Query, with SearchIssues' meaning (title or id contains the text, case-insensitively), on every Reader implementation, and issueops.SearchListRequest is search's scope as a request. Every bd search route (direct, --proxied-server and the HTTP backend) now asks Reader.List with it through one flag mapping; against a server that does not advertise the token the HTTP client refuses before dialing with a typed capability error. Two things a script can see follow from that mapping: bd search --type expands aliases (feat, mol, dec, ...) and validates the type as bd list does, so an unknown type now exits 1 where it used to answer an empty result, and bd search --json rows carry comments_omitted (additive) as bd list --json rows do.

  • bd create --graph now plans its batch through issueops.BatchApplier instead of the old buildDomainGraphPlan path, so a graph create gets the same atomic multi-row semantics as bd batch apply. A waits-for edge's metadata.spawner_id is stamped only when that edge names a spawner (never defaulted onto an edge that names none), and thread_id now survives translation into the batch request alongside every other GraphApplyEdge/GraphApplyNodeDep field — pinned by a reflection-based field-coverage test (TestBuildGraphApplyBatchRequestFieldsSurvive) that fails closed if a future field is added to either struct without an explicit decision about where it goes. The hierarchy/cycle rejection scenarios this path's local gates cover (blocking through existing or planned hierarchy, transitive external-parent paths, reverse parent-to-child edges, cycles hidden in an inline dep, combined scheduling cycles, the end gate alone with the per-edge cycle probe skipped) are exercised against a real Dolt store, not just skipped placeholders. A graph plan whose translated BatchApplier items (nodes, edges and deferred assignments combined) exceed issueops.MaxApplyBatchItems (1000) is now refused outright with GraphApplyTooLargeError rather than silently chunked across several transactions — bd create --graph promises one atomic request, so a plan over the cap must be split by the caller into multiple bd create --graph calls instead.

  • --if-revision (gastownhall/beads#4682) is extended to bd reopen, the one lifecycle verb #7203 did not add it to, reusing the same issueops.ReopenRequest.ExpectedVersion field the library already exposed and the same parseIfRevisionFlag / requireSingleIfRevisionID / reportIfRevisionFailure helpers #7203 introduced for the other four verbs — no new CLI surface, no new JSON body shape. A guarded reopen of an issue that is already open is still judged against the guard on both routes: a stale revision exits 13, and a matching one is the usual "already open" no-op. A faithful port of gc's (gascity) bdstore_conditional.go decode logic is now run against a real built bd for all five guarded verbs (TestEmbeddedGCConditionalMatcherDecode), confirming gc's matcher decodes #7203's numeric-only expected_revision/current_revision body exactly — no decimal-string twin fields are needed, since encoding/json decodes a JSON number straight into an int64 struct field with no lossy float64 intermediate. The proxied-server route (*_proxied_server.go's uow-backed preflight/apply path), previously untested for --if-revision on any verb, now has coverage for all five.

  • backends.Backend gains an optional OpenWith(ctx, beadsDir, OpenOptions) and a Remote bool field for a registered extension backend (for example an HTTP client registrant). OpenOptions{Credential, HTTPClient, UserAgent} carries per-open injections — the motivating case is a single embedder process serving many workspaces with distinct credentials against a backend whose dialer would otherwise be process-global. Open keeps working unmodified: a backend with no OpenWith falls back to it and ignores HTTPClient/UserAgent, but refuses rather than silently drops a non-nil Credential. The public SDK gains beads.OpenBestAvailableWith (and the backend package's matching aliases); beads.OpenBestAvailable is now that function called with a zero OpenOptions, with identical behavior for every existing backend. doltserver.ResolvePhysicalRoots now recognizes a registered remote backend (backends.IsRemote) before any Dolt-mode check, so a remote-backend workspace is never misclassified as having a local Dolt root to gate.

  • bd serve's GET /v0/beads/context now reports wire_revision (the non-additive wire-shape counter for this build) and min_client_wire_revision (the oldest revision this build still answers correctly). A client may declare the revision it was built for on any request via the optional Bd-Wire-Revision header (a plain non-negative decimal — no leading + or leading zero); a declared revision below min_client_wire_revision is refused with 400 invalid_argument/reason: "wire_revision_unsupported" and min_wire_revision, wire_revision, and bd_version fields naming this server's floor, current revision, and release string. The header is absent by default and an absent header is served exactly as before — this is additive, not a new precondition on existing clients. A client with no wire_revision signal to read yet may infer one from bd_version only as an ADVISORY hint (string-or-number, never a refusal on an inferred 0) — see the wire_revision property's inference notes in internal/httpapi/spec/openapi.v0.yaml for the cutoff and known-exception builds. See internal/httpapi/wire_revision.go and that same property for the full revision history.

  • GET /v0/beads/issues's sort parameter now advertises the issues.list.sort capability token, the same way an operation itself does, and a CI rule (TestNewParameterOnExistingOperationHasABehaviorToken) requires every new parameter on an existing operation, or new request body member on an existing operation, to carry a token that Capabilities() actually advertises (not just backticked a.b text shaped like one). Three frozen baselines (internal/httpapi/testdata/pretoken_*.json) grandfather what predates the rule and are pinned never to grow (TestPretokenBaselinesNeverGrow); a brand-new operation is permanently exempt — its own review decides its capability story.

  • A CI golden digest (internal/httpapi/wireshape) pins the JSON name, type, format, enum, required-ness, and nullability of every EXISTING response AND request-body member across the whole HTTP spec — including array item shape and additionalProperties value shape — recursing through $ref and through a schema's own allOf or oneOf, and fails if any changes without wire_revision bumping to match; go run ./internal/httpapi/wireshape/cmd/gendigest refuses to write a changed or removed entry unless wire_revision has moved past what the existing golden recorded, and refuses any write at a lower wire_revision (purely additive entries otherwise always write). The same digest also pins every operation PARAMETER (query, path and header), keyed by operationId + location + name, and records its type, item shape, enum, required-ness, effective style and explode (OpenAPI's defaults filled in where unset), and default — a parameter retyped, re-enumerated, narrowed, re-serialized, switched required, or removed is non-additive exactly like a response or request-body member. A new member or parameter always counts as additive, even a REQUIRED one an old client will not send, and value constraints (maxLength, pattern and the like) are outside the digest: such changes need their own review against wire_revision.

  • cmd/bd registers the http client backend built on top of the v0 bd serve wire (internal/httpclient), so a workspace whose .beads/metadata.json selects "backend": "http" opens over HTTP through the ordinary OpenBestAvailable path, exactly like a registered extension backend does. A new bd connect <url> command performs the handshake (api_version, wire revision, and — with --expect-project-id — workspace identity) and writes nothing until it succeeds; it then records the per-user activation sidecar (.beads/http_target.json, never git-tracked — cmd/bd/doctor/gitignore.go now requires it and its local metadata sidecar http_local_metadata.json be ignored) and sets metadata.json's backend to "http". bd connect refuses a plain http:// URL to a non-loopback host (a bearer credential would cross the network in the clear) unless --allow-plaintext is given, and refuses to switch a workspace that already selects a different backend unless --force. A credential is never accepted on the command line or written to disk; it comes from the same ladder every http request already uses (BEADS_HTTP_TOKEN=host[:port]=<token>, then BEADS_HTTP_TOKEN_COMMAND=host[:port]=<command>, then the credentials file, then none); both variables take only that host-scoped form, and a bare value is refused, since it would be sent to whatever server the workspace's http_target.json names. bd create sends no created_by on any backend: the create role defaults it to the request's actor (see Changed). The external-dependency policy decorator (internal/storage/externaldeps) still wraps an http store, and passes every ready, claim, claim-next, close, blocking and tree request through untouched when the server's handshake advertises policy.external_dependencies, which every bd serve composed through the policy now does (see below). Only against a server that does not advertise it (an older bd serve, or one handed raw roles) does the client apply the policy itself, in role terms: ready reads and the ready claim exclude the workspace-wide externally blocked set over a widened window, and per-issue checks (claim, close, batch close, list annotations, tree leaves) read only the named issues' edges. bd close --claim-next refuses over http before anything closes, since the wire's batch close cannot carry the claim; its error says to close without it and then run bd ready --claim.

  • bd serve advertises the conditional behavior token policy.external_dependencies exactly when the roles or provider it serves were composed through the external-dependency policy (httpapi.Config.ExternalDependencyPolicy, which bd serve derives with externaldeps.Applied / externaldeps.AppliedToProvider); a server handed raw roles withholds it. A client that sees it hands claim-next to the server as one atomic call and takes the ready count from the server, instead of judging external dependencies a second time against the CLIENT machine's project config.

  • bd ping works on an http workspace (a new storage.Pinger, which the http store answers with its authenticated context read).

  • The public backend/http package (bdhttp) is the out-of-tree door onto this backend for an embedder that links beads as a library rather than running cmd/bd: Register(Options) adds "http" to the registry and installs its transport, Open/Handshake dial a server directly with no workspace on disk, and the existing public beads.OpenBestAvailableWith takes a per-call OpenOptions.Credential through it — the case a multi-tenant embedder serving many workspaces needs, where a process-global credential cannot stand in for one tenant's own. EXPERIMENTAL, pin an exact beads version.

  • Unlimited ready and query reads over the HTTP backend (bd ready --limit 0, bd query --limit 0, and every library read asking for no limit) return every row against any server, bounded on the client by the new BEADS_HTTP_UNLIMITED_READ_CAP (a positive integer up to 1000000; default 10000). The client never puts limit=0 on the wire, which a bd serve --allow-non-loopback refuses and which an older server answered with its default page, silently capping the result: an unlimited ready or query read is one request bounded at the cap plus one (one consistent snapshot). Past the cap, or when a server reports more rows below the requested bound, the read is refused with a typed error (httpclient.ErrUnlimitedReadCap) naming the cap, never truncated. An unlimited bd list is unchanged: it walks its keyset cursor a page at a time and is bounded only by --max-rows / BEADS_MAX_ROWS. Unlike a malformed BEADS_MAX_ROWS, which is ignored with a warning, a malformed cap value fails every unlimited ready or query read rather than falling back to the default.

  • Two molecule capability tokens on bd serve. issues.close.autoCloseMolecule announces that POST /v0/beads/issues/{id}:close and POST /v0/beads/issues:batchClose accept auto_close_molecule and answer auto_closed_molecule and molecule_auto_close_refusal; issues.advanceMolecule is the new POST /v0/beads/issues/{id}:advanceMolecule, bd close --continue's advance. Both run the one library entry (issueops.CloseRequest.AutoCloseMolecule, issueops.MoleculeStepper) on the server. The public issueops package gains the molecule rules as one implementation over the BatchGetter, Relations and EdgeReader roles (MoleculeRoots, LoadMolecule, AnalyzeMolecule, ViewMolecule, ReadMoleculeProgress, ActiveMoleculeIDs, CompletedMolecule, MoleculeRootClose); a molecule's membership is the parent-child edge. Against a server that does not advertise a token the HTTP client refuses a library close that sets AutoCloseMolecule, or an advance, before dialing, with a typed capability error; it never runs a client-side copy of either rule.

  • bd close --json reports the molecule auto-close. The entry of a close that completed a molecule carries the root it closed as auto_closed_molecule, or why the root stayed open as molecule_auto_close_refusal, on every route; before, that refusal reached only stderr. Both members are absent from every other entry, so each entry still decodes as the closed issue, and the output stays an array (under closed with --suggest-next, --continue or --claim-next).

Changed

  • Over an HTTP workspace, bd ready --json without BD_JSON_ENVELOPE no longer sends GET /v0/beads/ready:count after a page the limit cut short: the total only went into the stderr hint. It lists one row past the limit instead, and the hint says more issues matched without saying how many (the wording the proxied-server route already uses). A local store, and the envelope's pagination.total, keep the total.

  • bd initializes its packages with GOMAXPROCS capped at 4 and hands it back before main runs anything, so the garbage collections package initialization triggers do not stop the world across every CPU of a large host. On a 192-CPU machine this takes about 17 ms off every bd process. An explicit GOMAXPROCS is restored as set.

  • Over an HTTP workspace, bd ready no longer fetches GET /v0/beads/context before its listing in every process only to learn whether the server enforces the external-dependency policy. The answer is kept for 5 minutes per server URL and pinned project under the user cache directory (BEADS_HTTP_POLICY_CACHE_TTL, a duration, a number of seconds, or off), and a handshake the process already holds still wins. Claims and closes always ask the server.

  • bd serve normalizes every RFC 3339 time bound on the issue listing and count (created_*, updated_*, closed_*, and the new defer_* and due_*) to UTC before it reaches storage. A bound carrying an offset (...+02:00) now selects exactly what its Z twin does; previously a backend whose timestamp column has no zone could read the offset bound's wall clock and compare it hours off. bd's own client always sent Z, so its answers are unchanged.

  • bd serve answers a label_regex the database cannot compile with 400 invalid_argument (invalid_value on label_regex) on GET /v0/beads/issues, GET /v0/beads/ready, GET /v0/beads/ready:count and POST /v0/beads/issues:claimNext, instead of 500 internal. The pattern is still compiled by the database in its own dialect; only the attribution of its refusal changed.

  • issueops.Reader.List refuses PinnedFlag together with NoPinnedFlag with ErrValidation (--pinned and --no-pinned are mutually exclusive) on every backend; it used to let PinnedFlag win silently. bd list already refused the pair at flag parse, so its behavior is unchanged; over HTTP the pair is a 400 invalid_argument on no_pinned.

  • issueops.Reader.List refuses ParentID together with NoParent with ErrValidation (--parent and --no-parent are mutually exclusive) on every backend, as CountRequest already did; it used to answer with only the edge-less rows under the parent's dotted id, which both clauses match. bd list already refused the pair at flag parse, so its behavior is unchanged; over HTTP the pair is a 400 invalid_argument on no_parent.

  • bd gate discover skips template gates, as bd gate check does. A template gh:run gate is a proto that work is poured from; on the direct, proxied and embedded routes bd gate discover used to record a discovered run ID on it, which every later pour copied. It is now left alone on every route. bd gate add-waiter on a template gate is refused by the update role's template guard, where the direct, proxied and embedded routes all used to write the waiter.

  • bd gate create refuses a negative --timeout on every route (timeout cannot be negative); the direct, proxied and embedded routes used to store it. A --type longer than 32 characters fails with the same typed validation error instead of a raw Dolt column error.

  • bd gate create and bd gate add-waiter follow dolt.auto-commit like every other role write. Under dolt.auto-commit=off bd gate create no longer makes its own Dolt commit on the direct and embedded routes (the proxied route already did not). The Dolt commit messages are the role's: bd: create gate X blocking Y becomes the create's default message, and bd gate add-waiter's becomes the update's (bd: update X on the direct and embedded routes, update issue on the proxied route). On the proxied route bd gate create no longer journals a dependency_added event on the blocked issue, matching the direct route and bd create --deps.

  • Gate members are bounds-checked in the library, on every local backend. issueops.Lifecycle.Create, Lifecycle.Update and BatchApplier (create and update items) now refuse, with a typed *issueops.GateFieldError wrapping ErrValidation: a negative Timeout (previously accepted and stored), an await_type over 32 characters, an await_id over 255 characters, and a waiter list whose JSON encoding exceeds its TEXT column (65535 bytes). Creates check all four; an update can change only await_id and waiters, so it checks those two. The over-width values previously failed with a raw Dolt driver error. There is no await_type vocabulary, no gate-type-only rule and no waiter deduplication.

  • Over an http workspace, a bd command answers its repeated reads of a row or setting from one response. bd show <id> --json makes one request instead of three, text bd show three instead of six, and bd update, bd close, bd label add, bd comments and bd dep add one to seven fewer (port of bd-enterprise e626072b8). A one-shot command's transport answers an identical getIssue, getSetting or listSettings GET from the answer it already received, for at most 5 s from that answer. Any write the command makes clears it, and only a 2xx that decoded is kept. bd serve, --watch and --follow never memoize, and neither do library embedders. Within one command a re-read no longer sees another writer's change made in the meantime (ledger row L-read-memo); BD_HTTP_READ_MEMO=0 turns the memo off. The identity-gate preview and the real open now share one transport, so a write makes one handshake. bd list reads its vocabulary with one listSettings, and the hook decorator skips its post-write re-reads when no hook for the event is installed.

  • created_by defaults to the actor in the library, for every create shape and every backend. PreparePublicCreateRequest, which Lifecycle.Create, BatchCreator items and batch-apply create items all prepare through, now sets an empty Issue.CreatedBy to the request's Actor; an explicit value is kept. The CLI (bd create, the state event create) and the three bd serve create handlers no longer stamp it themselves. A library caller that left CreatedBy empty used to store "" locally and the actor over http; both now store the actor. The raw CreateIssue seam is unchanged: import keeps the creator its rows carry, and bd swarm, bd todo, bd gate and the create form still stamp currentActor() there.

  • On an http workspace, bd update -s closed does not record closed_by_session. The update wire cannot carry it, so the CLI drops it rather than refusing the close; the status change, its close-policy guards and every other field in the same update still apply. A one-line notice on stderr names the drop, both for an explicit --session and for the ambient CLAUDE_SESSION_ID every agent under Claude Code exports, so it is never silent. bd close (and bd close --session) still records the session over http: its wire operation carries it. Local backends are unchanged.

  • On an http workspace, bd reopen sends no Provenance label. The CLI's label is the fixed bd: reopen <id>, which the wire cannot carry; the server records its own reopen entry for the same issue and actor, so the reopen no longer refuses. Local backends still record the label.

  • Routing settings are local-only on a remote backend. For an http workspace, routing.mode, routing.contributor, routing.default, routing.maintainer, contributor.auto_route and contributor.planning_repo come from local config (config.yaml or the environment) only; the server's shared settings document is not read for them, because routing targets are local filesystem paths that one shared document cannot name for every client. bd create's routing decision, bd migrate-personal and the missing-id lookup fallback all read the same local-only answer, and none pays a settings round trip for it. Local backends still fall back to the database values.

  • The molecule auto-close runs inside the step's close, and an error in it fails the close. bd close of a molecule's last step closes the root in the same transaction (opt-in AutoCloseMolecule, so an orchestrator that runs its own molecule lifecycle is not doubled), and it replays on an idempotent re-close. The root close is unforced, so a root a direct close would refuse stays open and is reported while the step's close stands: one its close policy refuses (such as a live blocker), and now also one a close guard refuses (a template, a pinned root, a root another actor holds). Any other failure of the root close (such as a storage error) now fails the step's close with it, where the CLI used to close the root afterwards and only warn. Two closes that finish a molecule's last steps at the same time no longer leave its root open: one of them retries and closes it. The cost is that concurrent closes of steps under one open auto-closing root now serialize, whether or not either finishes the molecule: each conflict is retried, and a close still conflicting when its retry budget runs out fails with the serialization error. A batch close reports each root once, on the first item under it that closed.

  • bd close --continue claims the next step for real. The advance is the MoleculeStepper role on every route: the assignee is set and a step another agent holds is skipped, never taken (the direct and http routes used to flip the status only).

  • bd mol current, bd mol progress, bd ready --mol and bd mol show read through the library's molecule entries on every route, so membership is the parent-child edge everywhere (bd mol show no longer adds children found only by an id pattern).

  • Against a server without the molecule tokens, bd close says what it did not do. If the server does not advertise issues.close.autoCloseMolecule (an older bd serve, or one whose orchestrator owns the molecule lifecycle), the close goes out without the auto-close request. When that leaves a molecule with every step closed and its root open, a notice on stderr names the root, once per root in a batch and again on a re-close; a step under an ordinary epic or an unfinished molecule prints nothing. If it does not advertise issues.advanceMolecule, or the handshake that would say so fails, bd close --continue refuses before closing anything, so the step is never left closed with the next step unclaimed.

  • Every bd invocation used to start two git subprocesses before doing any work: git rev-parse to locate the repository and git config user.name to resolve the actor. On Linux the repository is now located in-process (ordinary repositories, linked worktrees, submodules, no repository), producing exactly what git rev-parse printed and still running git for anything it does not model (GIT_DIR and related overrides, core.worktree/core.bare, repository extensions, includes, foreign ownership subject to safe.directory). The git config user.name actor fallback is resolved only when a command actually needs an actor, so read-only commands such as bd list and bd show start no git process for it. The actor's value and priority order are unchanged. In an embedded workspace with backup.enabled unset, the per-command auto-backup check ("is there a git remote?") is likewise answered from the git config files when they are plain (no includes, no command-line config), instead of running git -C .beads rev-parse --git-common-dir and git remote.

  • bd create --graph now stores a plan whose path from a node's parent to the node runs through a waits-for edge; the graph-only preflight that walked every ready-work edge used to refuse it. The plan now goes through the same end gate as bd dep add and bd batch apply, which walks only the scheduling edges (blocks, conditional-blocks, parent-child) and stores this shape too. Avoid it in plans: once the waits-for spawner has an open child, every issue in the shape, that child included, stays blocked and none is ever ready. The same path through scheduling edges alone is still refused.

  • bd preflight --fix --json no longer returns a Version sync fix result: version updates must keep all release surfaces aligned via scripts/update-versions.sh.

  • Release-tag pushes require Go and reject batches containing different release versions.

  • On version drift the release-tag pre-push hook points at the checker's remedies instead of prescribing scripts/update-versions.sh <version>, and scripts/check-versions.sh and bd preflight list the three that work: with cmd/bd/version.go already at the release version, that re-run rewrites only the .githooks markers and uv.lock and leaves any other drifted file as it was.

  • BREAKING (fail-closed): bd no longer migrates an existing database without explicit consent (Beads 1.2 release remediation). The 1.2.0/1.2.1 releases applied migrations 54–65 at store open on ANY command — including pure reads like bd list — with no prompt, and older binaries then refuse the migrated database (schema-skew guard). A command that would apply pending main-sequence migrations to an existing database at store open (in embedded mode that includes reads) now refuses with guidance instead, and the refused open performs ZERO writes (dolt_ignore seeding and clone-local migrations included). Server-mode read commands open without running migrations, so they keep reading at the current schema. bd dolt commit and bd vc commit warn and commit at the current schema without migrating, so the dirty-table recovery (commit, then bd migrate schema) still works on a database that is both behind and dirty. Consent is any of: bd migrate schema, the verb that names the migration (a preview flag withholds it); --force on bd migrate or bd migrate schema; or BD_ALLOW_MIGRATE=1 for scripts/CI (BD_ALLOW_REMOTE_MIGRATE=1 is honored as an alias). Bare bd migrate, including --update-repo-id, --inspect and --dry-run, does not consent. Fresh databases (bd init) still migrate to latest — creating a database is consent for its schema, kept across the retry of an interrupted server bootstrap — and the same creation principle covers a workspace the invocation just cloned into existence (bd init --remote, bd bootstrap sync), so the #4259 behind-remote flows keep their exact behavior. A database already at this binary's latest version never consults the gate; one left at v65 by 1.2.x or at v66 by 1.3.x still has migrations pending under this release, so it asks once. The remote-backed case keeps all of the #4259 remote-migrate gate's additional protections (smart gate, adopt fast-forward, fork-skew analysis) unchanged. JSON consumers: the refusal renders as a structured error whose error states the refusal and whose hint is a non-runnable operator directive (neither names a migration command or a consent variable), plus a migrate_consent object (current_version, required_version, pending, severity: "blocking", human_decision_required: true, and options: migrate runs bd migrate schema, or bd migrate schema --global under --global; keep has no command).

  • Shipped migrations are frozen by CI (same remediation): the content hash of every released up-migration (main 0001–0066, ignored 0001–0026, through the 1.3.x releases) is pinned by TestShippedMigrationsAreFrozen. Any amendment to a released migration must land as a new migration (0070+, or ignored 0028+, as of this change) so installed schemas and freshly-migrated ones cannot fork.

  • BREAKING: the close guards now hold for every close operation, not only bd close. The template read-only guard, the pin guard and the assignee authority fence (be-035) used to be a pre-read in cmd/bd, so a close through bd serve (POST /v0/beads/issues/{id}:close), issues:batchClose, a close item of issues:batchApply, or a library caller of issueops.Lifecycle.Close, BatchCloser or BatchApplier closed a template, a pinned issue, or a bead another actor holds that bd close refused. They now live in the role, inside the close's own transaction, on every backend: a template refuses with *issueops.TemplateReadOnlyError (no bypass), and unless Force is set a pinned issue refuses with *issueops.PinnedError and another actor's bead with *issueops.CloseNotAssigneeError (matching ErrNotOwner). Over HTTP they are 409 template_read_only, issue_pinned and not_assignee (the holder in assignee, also on a batch-close outcome), and the client rebuilds the same typed errors. bd close prints the same lines and exits as before, and bd close --if-revision, which skipped the guards, now applies them too. Two paths stay outside the guards. The raw storage CloseIssue, which a close line of bd batch and the other closes cmd/bd makes outside bd close use, runs neither the guards nor close policy on dolt, embedded or proxied; over HTTP it now sends force, so there it also stops applying close policy, and only a template refuses. A status update into the done category (bd update -s closed, a PATCH whose status is done, an update item of issues:batchApply) answers to close policy only, as before, so it still closes a pinned issue or another actor's bead. Migration: HTTP and library callers that close pinned issues or beads another actor holds must send force (Force on the request), or reclaim the bead first; the close operations never close a template, forced or not (pour it instead).

  • BREAKING (out-of-tree storage backends): a create without a priority now arrives as priority 0 with DefaultPriority set. bd create without --priority, a bd create --graph node without priority and a bd create --file template without a valid ### Priority used to reach a backend's IssueLifecycle().Create, BatchCreator().CreateBatch and BatchApplier().ApplyBatch as priority 2. They now arrive as priority 0 with issueops.CreateRequest.DefaultPriority (or BatchCreateItem / CreateItem.DefaultPriority) set, as does an HTTP create without priority when bd serve fronts the backend. The fields are additive, so such a backend compiles unchanged, but the shared preparation that applies the default lives under internal/, so a backend that ignores the flag now stores P0 (critical) for those bd create paths, and still stores P0 for that HTTP create. Migration: store issueops.DefaultCreatePriority when DefaultPriority is set, and refuse DefaultPriority with a non-zero priority as ErrValidation; conformance.RunRoleContracts checks both through RunLifecycleCreateAppliesTheDefaultPriority, RunBatchCreatorAppliesTheDefaultPriority and RunBatchApplyAppliesTheDefaultPriority.

Fixed

  • An HTTP workspace whose server cannot be reached now says which step failed. A connect that is refused names the address and says nothing is listening; one that times out (a filtered port, a down host) names the address that never answered and how long it waited; a host name that does not resolve is named as such. The error was the transport's Get "https://…/v0/beads/context": dial tcp …: i/o timeout, which named the request rather than the address.

  • bd link works over a connected HTTP workspace. It wrote through the store's raw AddDependencyWithOptions, which the HTTP client refuses, so every bd link failed there with operation "AddDependencyWithOptions" not supported by the http backend, whatever the type. It now writes through the DependencyEditor role, the route bd dep add takes, on the direct and HTTP routes alike (the proxied-server route is unchanged). Over HTTP that is one dependencies:add request, so bd link keeps accepting any non-empty type of at most 32 characters (owned-by, ships-in, ...). An empty or over-long type is still refused before anything is written, and an edge that already exists with another type still reports the typed conflict. On a Dolt store the history commit is now the role's, like bd dep add's: bd: dep add <from> <to> in place of dependency: add <type> <from> -> <to>.

  • bd query reports an uppercase duration unit as an unparseable time, not a negated internal error. created>30M failed with not a compact duration: "-30M", naming a value nobody typed; it now fails with cannot parse time expression: "30M". A signed spelling (created>+30M) stops at the parser, as created>+30x already did, so quote one to match it as a literal (label="+30M"). Units remain lowercase (30min, 6h, 7d, 2w, 3m, 1y); help and docs already say so.

  • Strict --readonly no longer creates workspace gate files. A bd --readonly command took its shared workspace gates with create-if-missing, so on a workspace no bd command had gated yet it left <project>/.beads.gate.lock and the physical-root gate file (.beads/embeddeddolt.gate.lock or .beads/dolt.gate.lock) behind, although strict readonly promises not to mutate the workspace. It now opens an existing gate file read-only and holds it shared as before, so a maintenance operation still sees it and a running one still makes it wait; when no gate file exists it holds nothing. Every other command still creates the files.

  • Strict --readonly refuses bd backup init, backup sync, backup remove, backup restore and bd import. None of them checked readonly mode. On a server-mode store they ran to completion under --readonly: backup init and backup remove rewrote or emptied the server's backup registry, backup sync wrote the backup destination, backup restore --force replaced the store, and import wrote issues. On an embedded store they failed with the storage layer's store is read-only error instead, after backup sync had left .beads/backup.lock behind and backup restore had left workspace-gate holder files. They now exit 1 with the standard Error: operation '<op>' is not allowed in read-only mode, and a restore is refused before it takes the exclusive workspace gate. bd import --dry-run is refused too, as bd create --dry-run already is and as a migration freeze already refuses it. bd backup status still runs under --readonly.

  • bd init --server --external no longer starts a Dolt server of its own. Init builds its store config by hand and turned auto-start on for every server init, so an --external init that named no port launched a repo-local dolt sql-server, although --external says the server is managed outside bd. --external now turns auto-start off. When the external server cannot be reached, init gives one message on every route (host and port, socket, shared server, auto-start pinned off): it names the endpoint it dialed, says that bd did not start a server, and no longer advises bd dolt start. With no port named at all, it asks for one (--server-port or BEADS_DOLT_SERVER_PORT) instead of offering a default.

  • A failed bd init --server can be retried. A first attempt that could not reach its server left an empty .beads/dolt and nothing on disk that records the server choice, so the legacy-upgrade guard refused every retry as a "legacy Dolt workspace" and no bd command could repair it. When init itself selects server mode, the guard now admits exactly that root: empty, with no .local_version witness. A pre-1.0 witness, or a .beads/dolt that holds anything, is still refused, and commands other than bd init are decided as before.

  • A graph-mode workspace is refused instead of opened as an Issue store. Graph-preview builds mark their workspaces with "graph_mode": "link" and graph_* binding fields in metadata.json, beside an ordinary Dolt backend. bd opened such a workspace as an Issue store and ran Issue schema initialization against the graph database, and every metadata.json rewrite (including the config.json to metadata.json migration) dropped the marker and the binding. Now any graph_mode other than dependency, including a value bd does not recognize, makes every command that runs in the workspace or names it with --db or BEADS_DB (bd init, bd doctor and bd bootstrap included), and bd create --repo aimed at it, refuse it before touching anything. The refusal names the format and exits with code 16, so installed beads git hooks refuse commits and pushes there too. Commands that reach such a workspace only through a route from another one, such as a routed bd show, fail without touching it, but not with code 16. Store-less commands such as bd version and bd help still run. A metadata.json rewrite keeps graph_mode and the graph_* fields graph-preview builds write, and like those builds it omits a zero graph_schema_version and a false graph_ready.

  • A graph workspace without graph_mode in its metadata is refused too. Graph-preview builds write a .beads/graph-preview-format marker before metadata.json. bd ignored it. When an older bd had rewritten metadata.json without graph_mode, bd opened the graph database by its recorded name and ran Issue schema initialization against it; when metadata.json was never written or was lost, bd fell back to the default database name and created an Issue database inside the graph workspace, and bd init initialized an Issue workspace there. A .beads holding that marker, whatever it contains and whatever its metadata says (absent, stripped, unparseable or an explicit dependency), is now refused wherever a graph_mode workspace is: commands run in it or naming it with --db or BEADS_DB, bd create --repo aimed at it, and bd bootstrap in a workspace below it that would take its metadata exit with code 16 and a hint to restore metadata.json, and routed and cross-repository opens fail without opening its database.

  • The Go library's opens refuse a graph workspace too. beads.OpenBestAvailable, OpenBestAvailableWith, OpenFromConfig and OpenGated opened a workspace whose metadata.json declares a graph_mode other than dependency, or whose .beads holds the graph-preview-format marker, as an Issue store and ran Issue schema migrations into the graph database. They now return *beads.UnsupportedWorkspaceFormatError (match it with errors.As) before opening, creating or changing anything, the same refusal the CLI exits 16 with.

  • bd setup claude --remove no longer leaves a blank-line hole. Removing the beads section from CLAUDE.md (or, with the other setup recipes, from AGENTS.md or GEMINI.md) left the blank lines around it behind, three in the CLAUDE.md bd init writes, and setup then re-added the section after two more, so every remove and re-add grew the file. --remove now takes those blank lines with the section, and setup appends a section after one blank line, so removing and re-adding it toggles between the same two files. A re-added section goes at the end of the file.

  • bd config set no longer truncates a config.yaml that has a line of 64 KiB or more. Setting an undotted key (such as actor or no-db) rewrites the file line by line, and a line of 65,536 bytes or more ended that read early without an error, so the command reported success and wrote back only the lines before the long one, plus the key it set. Dotted keys and bd config unset were not affected. Setting an undotted key in such a file now fails with bufio.Scanner: token too long. bd config set leaves the file as it was. set-many stops at that key, but it has already written any config.yaml keys listed before it, such as a dotted sync.remote.

  • bd show --as-of exits 1 when it reads no issue. A ref the Dolt AS OF read does not take (an instant, a versioned-history address), a ref that does not exist, or an issue that did not exist at the ref was reported on stderr while the command exited 0 with nothing on stdout, even under --json. It now ends the way plain bd show does: exit 1, with the JSON error envelope on stdout under --json. A batch that reads at least one of its ids still exits 0. Both the direct and the proxied-server routes changed.

  • bd list --format dot prints the same graph the same way every time. Nodes are declared in id order and edges in source, target and type order. They used to follow the listing order, which breaks a priority tie by creation time to the second, so two equal-priority issues swapped places depending on whether they were created in the same second, and the proxied-server route did not pin the order of one issue's edges. --sort and --limit still choose which issues are listed.

  • bd init --quiet no longer prints "Removed SessionStart hook" or "Removed PreCompact hook" when the project's .claude/settings.json (or the legacy .claude/settings.local.json) still carries the bare bd prime hooks an older bd registered. The hooks are still migrated to the SessionStart bd prime --hook-json hook; the removal lines now go to the installer's own output, which --quiet discards, instead of straight to stdout.

  • A new AGENTS.md from bd init lists bd dolt push in its Quick Reference only when the workspace has a Dolt remote. The managed Beads section already left out its push lines when no Dolt remote is configured or the workspace is no-push, but the Quick Reference above it still showed bd dolt push # Push beads data to remote. It now follows the same rule. bd init still leaves an existing AGENTS.md alone outside the managed section, and writes an --agents-template as given.

  • bd setup claude no longer rewrites a CLAUDE.md whose beads section is already current. A rerun (and bd init, which runs it) wrote the same bytes back as a new file, resetting its mode to 0644 and its mtime, and reported "Updated existing beads section"; it now leaves the file alone and reports the section current (the report bd init already gives a current AGENTS.md). bd setup factory, opencode, mux and gemini share the fix. A section whose END marker is missing or comes before BEGIN, which was also reported as updated while nothing changed, is now reported as an error: bd setup factory, opencode and mux exit 1 (mux before installing its .mux hooks), while claude and gemini warn and carry on. When CLAUDE.md is an @AGENTS.md stub and it is AGENTS.md that cannot be updated, claude now names AGENTS.md in the warning and keeps the beads block left in the stub, which it used to delete although AGENTS.md was never updated.

  • Time-bound filters and the keyset cursor compare the same instant on every engine, whatever zone they are given in. The storage filter builder binds every time bound (created_*, updated_*, closed_*, started_*, defer_*, due_*) as the UTC spelling of its instant. An engine whose timestamp column carries no zone drops a literal's offset, so a +02:00 bound would have compared as that wall clock; Dolt already honored the offset and bd resolves bounds to UTC, so no answer changes here. The keyset cursor (AfterCreatedAt) is now bound in UTC too: embedded Dolt's driver sends a time.Time as its wall clock with no zone, so a library caller paging with a cursor in another zone was answered from the wrong position there (server Dolt was unaffected).

  • bd setup claude --global keeps its beads section in ~/.claude/CLAUDE.md. It wrote the section into the current directory's CLAUDE.md: run from your home directory it created ~/CLAUDE.md, which by default stops Claude Code reading AGENTS.md in the projects below it, and --remove --global stripped the section a project's own install had put in its CLAUDE.md while leaving that project's hooks in place. A global install and removal now use only the user-level files beside ~/.claude/settings.json. A new ~/.claude/CLAUDE.md holds just the section, rendered the same whichever project you run from. Claude Code loads that file in every project, so its section, task-tracking rules included, applies only in repositories that use beads. With global or plugin hooks --check accepts it when the project's instructions have none, or names --global when it is missing and the project has no CLAUDE.md of its own. Run from ~/.claude, or from your home directory unless ~/CLAUDE.md still holds a section an earlier run left there, --check judges the user-level files, and from your home directory it names --global when no hooks are installed. From your home directory, whose .claude/settings.json is the global one, --remove removes the user-level section with the hooks. If you ran --global before, re-run bd setup claude --global to add the section to ~/.claude/CLAUDE.md. A section an earlier --global run left in a directory's CLAUDE.md stays there; delete it by hand if you don't want it.

  • bd blocked --label, --label-any and --exclude-label apply to a row blocked only by an unsatisfied external: dependency. The external-dependency policy lists such a row itself, after the label query has run, and it was listed whatever its labels; it is now narrowed by the label filters like every other row, on every route. On the proxied-server route such a row also printed with no labels; it now prints them.

  • bd gate add-waiter survives contention the store gave up replaying. It retried only a version miss, so a convoy adder failed when its guarded write ended in a serialization failure or deadlock the backend stopped replaying (Dolt/MySQL 1213/1205, Postgres 40001/40P01), or, over HTTP, in a 503 busy. Those are now retried, but at most twice: each arrives only after the store or bd serve has spent a whole replay budget on the write (up to 15s), so the 20-write bound a version miss gets could hold an adder for minutes. Each retry waits the jittered backoff plus a busy answer's Retry-After, as bd serve sent it. The retry is safe because every round re-reads the gate, deduplicates and writes the whole list guarded by the version it read: a 503 is not proof that nothing was written (a gateway can answer one after the commit), and a write that did land is found on the re-read, so the waiter is never added twice; when the last write allowed is answered busy, one more read checks for it before the command fails. A server fault (500) and an unreachable database are still returned at once, and once the retries run out the error says no write was confirmed and names the write contention, instead of saying the gate kept changing.

  • bd list --title-contains X works against an HTTP workspace. The client folds it onto the listing's existing title parameter (the store compiles both flags to the same case-insensitive substring predicate), so it needs the issues.list.title capability, like --title. A --title and a --title-contains that differ other than in letter case are two predicates, so they ride title and title_contains together and also need issues.list.filters (see Added), as does the --closed-after in gascity's polecat pre-flight lookup of recently closed beads. Either flag with --parent in the default tree view still refuses over HTTP.

  • bd gate create, bd gate add-waiter and bd gate discover work over a connected HTTP workspace. They wrote through the legacy CreateIssue, AddDependency and map-based UpdateIssue, which the HTTP store does not implement, so each failed before any gate member reached the wire. They now run through the library roles on every route:

    • bd gate create is ONE Lifecycle.Create carrying the gate members and a reverse blocks edge from the blocked issue to the new gate, so the gate and its edge land atomically (one createIssue over HTTP, behind issues.create.gates). The direct route used to make two writes and an explicit commit; the proxied-server route now uses the same role.
    • bd gate add-waiter reads the gate through Reader.Get, deduplicates, and writes the whole waiter list through Lifecycle.Update guarded by the row version it read. On a version mismatch it re-reads and retries after a jittered, doubling backoff (10 ms up to 400 ms per wait, at most 20 writes), so a convoy of workers registering on one gate at once all succeed. On the direct route concurrent adders used to lose each other's waiters while every one exited 0; no waiter is lost now. Over HTTP it needs issues.update.gates.
    • bd gate discover lists through issueops.ListGates (as bd gate check does) and records each discovered run ID through Lifecycle.Update, the same await_id write bd gate check and bd close make. Over HTTP it needs issues.update.gates. Its summary now counts only the gates whose run ID was recorded, and it exits non-zero when any matched gate's write failed (it used to report such a gate as updated and exit 0).

    Against a server that does not advertise the token it needs, bd gate create and bd gate add-waiter are refused before any write, naming the token; bd gate discover writes nothing and reports the token as that gate's error. The served corpus gains mc-gate-create-timer, mc-gate-check-timer-resolves, mc-gate-add-waiter and mc-update-await-id (pending until the reference server advertises the gate tokens), pinned passing by TestE2E_GatesOverHTTP.

  • bd import no longer reports a failed dependency check as a skipped dependency. Import skips an edge the dependency checks refuse (a self-dependency, a cycle, or a blocker that is the issue's own ancestor or descendant) and reports it on a Skipped dependency: line, or in skipped_dependencies with --json. It did the same when the check's own query failed, for example on a timeout, a dropped connection or a statement the backend could not run: the SQL error became the skip reason, and on Dolt, where the transaction survives a failed statement, the import committed without the edge and exited 0. Such an error now fails the import. Refused edges are still skipped as before.

  • bd mol bond <A> <B> ignored prefix routing — operand resolution now uses the same routing fallback as bd show/bd update/bd close, so a bead that lives in a routed store (a prefix-routed rig via routes.jsonl, or the contributor planning store) resolves instead of failing with '<id>' not found as issue or formula. Previously bd mol bond resolved operands against the local store only, so gt sling <bead> <rig> — which cooks the default formula and bonds it to a rig-routed bead — died at bd mol bond even though bd show <bead> resolved the same ID. Resolution is write-intent: the routed target opens writable and the bond commits where the bead lives. Two operands that live in the same routed rig share one store handle and bond normally; only a bond whose operands resolve to genuinely different databases is rejected rather than written to the wrong one. As part of this change, an auto-route — contributor, maintainer or default — now opens writable for the commands that change an existing issue (bd update, bd close, bd comment, bd unclaim, bd dep and the other routed mutation commands, as well as bd mol bond), so the write commits where the issue lives: for a contributor, in the planning repository, as the contributor workflow documents. Every auto-route used to open read-only, so on an embedded workspace these writes were refused. bd unclaim now opens a routed issue writable, as bd update --claim does, and commits the release there; on an embedded workspace bd human respond and bd human dismiss now record a routed change as a Dolt commit in the routed repository instead of leaving it uncommitted in that repository's working set. A routed write, prefix or auto, never creates the target's database: an embedded repository that has not been bootstrapped yet fails the write instead of gaining an empty database that bd bootstrap would then take for the real one. An auto-routed store, read or write, now opens the target's own configured dolt_database, as a prefix route already did, even when BEADS_DOLT_SERVER_DATABASE names the current workspace's database. An auto-routed write refuses a target whose metadata.json is unreadable or names no dolt_database, while a target with no metadata.json at all (a planning repository nothing has been created in yet) opens the one embedded database it holds, or its default database when it holds none or several; a prefix route still requires the target's dolt_database. A routing failure, such as an unreadable routes.jsonl or a routed store that fails to open, is now reported as itself instead of as no issue found matching. Same routing-parity fix as #3608 for bd close. Fixes #4714.

  • bd gate check works over a connected HTTP workspace. Its open-gate listing built the same raw SearchIssues filter bd gate list did before S20, so the HTTP backend refused the command before reading a gate and gc's core gate-sweep order (bd gate check --type=timer --escalate) failed on every remote scope. It now lists through issueops.ListGates over the route's Querier (direct, proxied and remote alike). Closing a resolved gate was already served. A bead gate's first sighting of its awaited bead (await_seen, recorded when the check finds that bead, not yet closed, in the gate's own store) was written through the legacy map-based UpdateIssue, which the HTTP store does not implement, so every open bead gate with an open target was an error on every sweep; it now goes through the Lifecycle role's Update (a single-key metadata set, issues.update on the wire), as does the gh:run discovery's await_id persist, which the HTTP client sends to a server that advertises issues.update.gates (see Added). Against an older server it refuses that patch before the dial, so a gh:run gate with a workflow-name hint stays that gate's error, naming the missing token and the server rather than a Go method. The sweep also skips the gates of a persisted proto (bd cook --persist): they are templates that every pour copies, so a write to one was refused on every sweep and, in a local workspace, a resolved one (an expired timer, say) was closed in the proto itself. The listing's query leaves them out (template=false), so they take no row of --limit and cannot crowd a live gate out of the sweep; bd gate list still lists them. A negative --limit is now an error; the direct and proxied routes used to read it as no limit. The served corpus gains mc-gate-check-timer, mc-gate-check-gh, gate-check-all and a served test over await-typed and template gates.

  • bd list's text renderings read only the listed issues' dependency edges. The tree (the default), --deps, --parent, --format and --watch loaded every dependency record in the workspace (GetAllDependencyRecords) to draw one page. Over a connected HTTP workspace that is a walk of the whole workspace: every issue id, 200 per request, then every id's edges, 100 per request. bd list --limit 20 against a ~3,800-issue workspace made 62 sequential requests (~5.7 s at ~50 ms a round trip). Every rendering reads only edges a listed issue owns, so the listing now reads exactly those through the EdgeReader role: one request per 100 listed ids over HTTP. The same listing makes 6 requests (~0.5 s), one more than bd list --json, and prints the same bytes. On a local store a default page is as fast or faster. A listing that covers most of a large store now pays batched reads (four queries per 200 listed ids) instead of one scan of both edge tables, which costs more past roughly 1,000 listed rows: +0.07 s at 1,000 rows and +0.46 s at 4,000 on embedded Dolt. A store with many wisps, whose edges the old read also scanned, is faster at every page size.

  • A heartbeat can no longer resurrect a lease after a release or close. A heartbeat re-arming a hand-doled claim (in_progress with no lease row) could race a release or close on Dolt and leave an orphaned lease row on an open or closed issue; every later heartbeat by the former holder then refreshed it and answered success for a claim it no longer held. This affected every Dolt route, including releases before this one. The re-arm is now recorded as a mutation and remints the issue's row_lock, which every claim-ending path rewrites, so it collides with a racing release or close and one side is replayed against the other's result (the cost: a hand-doled claim's first heartbeat changes its revision once). The refresh also requires, in the same transaction, that the issue is still in_progress under the holder, and a refusal deletes the caller's orphaned lease row, so a lease row orphaned any other way never answers a heartbeat. A refresh and a refusal still make no Dolt commit on any route; a re-arm commits only its own tables (never DOLT_COMMIT('-Am')), so another writer's uncommitted batch-mode edits to other tables stay uncommitted (a re-arm stages its tables whole, as every other direct-route write does). On embedded Dolt, a heartbeat landing in the same second as the previous one no longer re-grants the lease (rewriting its granted_at / granted_node), and on every route a heartbeat whose actor is a different spelling of the lease's holder (gastown.mayor for gastown__mayor) refreshes the lease in place, keeping its granted_at and granting replica, instead of re-granting it under this node. For a Postgres port: the refresh now reads the issue row, so under SERIALIZABLE expect more retried serialization failures between a holder's own edits and its heartbeat.

  • bd gate list works over a connected HTTP workspace (S20). The workspace-wide listing built a raw SearchIssues filter, which the HTTP backend refuses, so gc's renudge-stale-human-gates order failed on every remote scope. Every route now calls one library entry, issueops.ListGates, which asks the Querier role (queryIssues, served by every deployed server) for type=gate with closed rows hidden unless --all: the same rows the raw filter selected, on both planes. --limit 0 is unlimited, as before. The bead-scoped form (bd gate list <id>) is unchanged.

  • bd export works over a connected HTTP workspace (S21). It read through the raw SearchIssues method and the bulk label, dependency and comment reads, which the HTTP backend refuses, so gc's 15-minute JSONL export order (gc bd export --all) failed on every remote scope. On a remote backend the read set now comes from one library entry, issueops.ReadExport, composed over role reads every deployed server publishes: an unlimited Reader.List for the rows, their labels and counts, EdgeReader.ReadEdges for the dependency records, and Reader.Get with comments for each row that has any. The JSONL is written by the same code as on every other route. Memories (--all, --include-memories) come from the served memory plane (listMemories), not the settings listing, which withholds every kv.* row and so exported none. A failed read fails the export, never shortens it. A remote export does not yet stamp the wisp_plane marker, because the server publishes no plane membership. So a current bd import of a remote export puts genuine no-history wisps on the durable issues plane instead of the wisps plane, and gc's JSONL archive keeps them. Nothing is dropped.

  • A silent HTTP connection no longer costs a read the full 60s. The HTTP backend client gives a read's first attempt 15s from when it is sent to get its response headers (BEADS_HTTP_RESPONSE_HEADER_TIMEOUT: a duration, a number of seconds, or off), then retries it once on another connection with no header timer, on what is left of the same 60s request timeout. A slow but healthy read still finishes if its retry's headers and body together fit in that remainder (about 45s); one that needs longer now fails at 60s. A read whose connection is reset is retried the same way (on Windows only when the reset surfaces as an early EOF), and each retry is logged to stderr with its count. The events:watch stream's connect gets neither the timer nor the retry. A write is never retried; one that was sent and timed out before its headers fails with a typed wire.ResponseHeaderTimeoutError naming the server's host, because it may or may not have been applied, while a dial or TLS timeout stays a plain connection error. The batch get, a read on POST, is now also retried on a 503 the way a GET is.

  • A large batch apply over HTTP no longer fails at 60s while the server commits it. bd serve runs an issues:batchApply of more than 100 items (a large bd create --graph plan, say) under its --large-apply-ceiling (5 minutes by default), on a context detached from the client, but the HTTP backend client gave every request 60s. A large apply that took longer failed client-side with a timeout while the server went on to commit it. Such a request is now dialed with 6 minutes, which covers the server's default ceiling plus what it may spend before (reading the request, waiting for the one-wide large-apply slot) and after (writing the answer back). Every other request keeps 60s. An *http.Client passed in through backend/http's Options.HTTPClient is raised the same way for that one request when its Timeout is shorter than 6 minutes; one with no Timeout stays unbounded. A server whose --large-apply-ceiling is raised past about 5m15s can still outlast the client: it does not advertise its ceiling, so the client cannot learn it.

  • An update of a template is now refused for every caller, not only bd update. The template guard used to be a pre-read in cmd/bd, so an update through bd serve (PATCH /v0/beads/issues/{id}), an update item of issues:batchApply, or a library caller of issueops.Lifecycle.Update or BatchApplier edited a template that bd update refused. It now lives in the role, in the update's own transaction, on every backend: any update naming a template refuses with *issueops.TemplateReadOnlyError (matching ErrTemplateReadOnly), whatever its patch and force flags, after the compare-and-set guards. Over HTTP it is 409 template_read_only, and the client rebuilds the same typed error. HTTP and library callers that edited templates are now refused (pour the template instead) unless they set the new UpdateRequest.AllowTemplate (allow_template on the PATCH body), which stands the guard down for that one request. bd update, bd assign and the proxied bd tag print the same line as before; bd label and bd set-state set AllowTemplate and keep editing templates as they always have. A batch create that makes a template and splices its metadata still lands: the splice finishes the create. Version skew: a new bd against an older bd serve (no issues.update.allowTemplate token in the handshake) refuses a single template update itself before dialing, and never sends allow_template to it. A batch update item gets no such check (its target resolves on the server), so through an older bd serve it still edits a template. An older bd label or bd set-state against a new bd serve cannot send allow_template, so it now gets 409 template_read_only on a template; upgrade the client. Not yet moved: bd comment, bd note, bd priority and bd tag still refuse templates with their own pre-read in cmd/bd (their direct-route writes do not go through Lifecycle.Update), so for those four the guard lives only in the CLI. Claim, reopen, :casMetadata and the served addComment have no template check at all yet (bd-jkp9v3): POST /v0/beads/issues/{id}:claim (issueops.Claimer) still claims a template that a PATCH with claim: true refuses. Close's guards move to the role in their own change (#7425).

  • The dotted-id hierarchy refusal (a child such as bd-abc.1 may not carry an explicit edge to its own ancestor, other than the parent-child edge to its immediate parent) moved from cmd/bd into the library as issueops.CheckDottedChildDependency / *issueops.DottedChildDependencyError (unwraps to ErrValidation). Every DependencyEditor (dolt, embedded, unit of work, HTTP) and every BatchApplier dep_add item now enforces it, so an HTTP dependencies:add, a batch:apply edge and bd create --graph's own edges are refused like bd dep add (HTTP: 400 invalid_argument). bd dep add, bd dep --blocks, bd link and bd dep add --file print the same messages as before, and a bd batch script's dep add line, which neither backend checked, is now refused with that message and rolls the whole batch back. The refusal is decided from the two ids, ahead of every check that reads the stored graph, so for a child whose parent-child edge is already stored (a bd create --parent child) an HTTP caller now gets that 400 where it used to get 409 dependency_cycle (a blocking type) or dependency_exists (any other type).

  • An HTTP create (issues:create, issues:batchCreate, a batch:apply create item) that omits priority now stores the create default, P2, as the spec documents, instead of P0 (critical). The default lives in one place: issueops.CreateRequest.DefaultPriority (and BatchCreateItem / CreateItem.DefaultPriority) asks the shared create preparation to store issueops.DefaultCreatePriority; the handlers set it for an absent member, the HTTP client omits priority for it only when the server advertises the new additive issues.create.defaultPriority handshake token (an older bd serve reads an absent priority as P0, so a new client sends 2 explicitly to it), and bd create (no --priority), bd create --graph (a node without priority) and bd create --file (a template without ### Priority) rely on it rather than spelling 2. An explicit 0 is still P0, and DefaultPriority with a non-zero priority is ErrValidation. CLI output is unchanged on the in-tree backends, except that bd create --file now warns on stderr about a ### Priority it cannot parse, as it already did for ### Type, rather than dropping it silently. The client decides from the handshake its store cached (once per command in bd), so a create that lands on an older build behind the same URL (a bd serve rolled back under a live store, or mixed builds during a rollout) still stores P0, with no error on either side. Out-of-tree storage backends must now honor DefaultPriority; see the BREAKING (out-of-tree storage backends) entry under ### Changed.

  • Blocked-state repair no longer lets a stale is_blocked bit on a parent with no reason row of its own hold itself up through a parent that blocks on its own child, in a malformed parent-child cycle (gastownhall/beads#7386). The explained-parent walk, on both the full repair and the per-write recompute, now climbs through such a parent, whose bit is inherited, instead of treating it as a final unexplained ancestor; bd doctor's Blocked State check counts the stale bit and --fix clears it.

  • bd show --current over the HTTP backend no longer prints the wrong issue. Its in-progress/hooked lookup used the raw SearchIssues method, which the HTTP backend refuses; the refusal was swallowed and the command fell back to the last-touched issue, exit 0. The lookup now reads through the issueops.Querier role, and a failed read is an error on every route (including --proxied-server), never a fallback. A miss (nothing in progress or hooked, or no actor) is another matter: --proxied-server now falls back to the last-touched issue as the direct route always has, where it used to fail with no current issue found.

  • bd list --type <infra type> --include-ephemeral (and GET /v0/beads/issues with the same parameters) now include that type's durable rows — a promoted wisp, an issue retyped after it was created — beside its wisps, as --include-ephemeral promises; the type used to narrow the read to its wisps alone. bd count --include-infra --include-ephemeral and the count endpoint count the same rows. An infra --type without --include-ephemeral still reads only the wisps, and bd search --type <infra type>, now served by the same listing, still finds that type's wisps and durable rows alike.

  • bd recompute-blocked on a workspace connected to bd serve is a no-op that exits 0 instead of refusing: the server maintains is_blocked on every write and the column lives in its database. --json reports {"rows_corrected": 0, "maintained_by": "server"}.

  • The http store's GetIssue answers a miss with storage.ErrNotFound, like the local stores. It used to answer a missing (or empty) id with (nil, nil), which every caller written against the backend contract read as a hit: bd create --graph refused every plan naming an explicit id over http. The miss is now not found: issue <id>, wrapping storage.ErrNotFound exactly as issueops.GetIssueInTx does; the empty id never dials.

  • bd list, bd children, bd show --json counts, the default bd dep tree, bd ready --claim and bd close work on an http workspace. The external-dependency decorator rebuilt its views from storage seams the http store only partly serves, so bd list/bd children refused, bd show --json printed zero dependency and comment counts without an error, and the default dep tree refused; each view now passes the server's own role through (see Added). A missing id is a plain not-found after one getIssue, instead of a partial-id search refusal, and bd update <absent> --claim answers a not-found gc classifies.

  • bd mol wisp gc --age 0s can miss a just-created wisp. Dolt's DATETIME(0) column rounds fractional seconds half-up on write rather than truncating, so a wisp created in the second half of a wall-clock second can read back with updated_at up to 0.5s in the future. The GC's age check (now.Sub(updatedAt) > ageThreshold) then compares against a negative duration and fails, even though --age 0s's own contract is that every wisp qualifies. ageThreshold <= 0 now bypasses the clock comparison entirely (#7236).

  • bd mol wisp gc no longer deletes a fresh step because a wisp it depends on is abandoned. The GC applied --age only to the wisps it selected directly, then deleted their dependents at any age as long as they were not live work, so a step created seconds earlier went with an old wisp it depended on, and raising --age did not help. A dependent now has to meet --age itself, by the same rule as the wisp it depends on (including the #7236 rounding fix). One that is too young is spared and loses its edge to the deleted wisp: an open one is collected once it ages out, a closed one is left for --all, --closed or bd purge, since without --all the GC reaches a closed wisp only as the dependent of an abandoned one (#5219).

  • bd create --parent works on the HTTP backend: the create role mints the <parent>.<n> child id (and inherits the parent's labels) inside the create on every route, instead of the CLI reserving it through a raw GetNextChildID the HTTP backend does not serve. bd mol current (with or without an id) and bd mol progress no longer answer empty or refuse over http, and bd ready --mol step rows carry their assignee and timestamps.

  • Server-mode config writes are committed again. On the direct SQL-server route, bd remember, bd forget and bd config set|unset|set-many wrote the config table but never created a Dolt commit: maybeAutoCommit returns early off the embedded route, and plain Commit() excludes config (GH#2455). The rows sat in the working set, where bd dolt push does not carry them, and a dirty internal key such as status.custom made the next bd dolt pull refuse until an explicit bd dolt commit. With dolt.auto-commit on (the default) these writes now commit right away through the scoped CommitConfigOnly, which stages only config and the lookup tables projected from it, so a concurrent operation's other dirty tables are never swept; batch and off defer them to bd dolt commit like every other write. bd remember and bd forget commit only user kv.* rows. When an internal config key is dirty (such as the issue_prefix that bd rename-prefix writes), the memory is still stored but not committed: the command warns on stderr with the keys named and exits 0, --json adds committed: false, commit_error and dirty_config_keys, and bd dolt commit publishes the memory with those keys. v1.0.1 (#3052), whose commits this restores, also warned on a failed memory commit instead of failing the write. bd config unset of a yaml-only secret key also commits its delete of a credential row an older bd stored in the database, so HEAD, which bd dolt push publishes, no longer keeps it (#4078).

  • PRs based on hotfix/** branches now run full CI, not just cross-version historical smokes and triage labeling. pr.yml, pr-risk.yml, conformance.yml, cross-version-smoke.yml and regression.yml all trigger pull_request on main and release/**, but not on hotfix/** — a backport PR based on a hotfix line (the same position release/** was added for) got no unit shards, no lint, no risk gate, and no required CI Gate at all: 15 passing cross-version smokes and triage jobs, nothing else. Added hotfix/** alongside release/** in all five (#7148).

  • Ordinary bd commands now wait out a maintenance operation instead of failing at once. While bd init, bd backup restore, bd migrate or bd bootstrap holds the workspace gate exclusively, every other bd command on that workspace — and, in shared-server mode, on every project sharing the server (another project's ~8s bd init) — failed immediately with "a maintenance operation is running". They now wait up to 30s (one "waiting for another bd process" notice after 2s; Ctrl-C aborts), then fail with the same error naming the bound (and, in shared-server mode, saying the holder may be another project). Override with BEADS_GATE_WAIT_TIMEOUT (1m, 90); scripts that relied on the old fail-fast behavior should set BEADS_GATE_WAIT_TIMEOUT=0. Git hooks stay fail-fast so a commit or checkout is never stalled. Waiting is writer-fair: once a maintenance operation is waiting for the gate, newly started commands queue behind it, so a steady stream of short commands can no longer starve bd init. The queue only ever delays a command, never fails it, and a maintenance operation that cannot get in (for example behind a bd list --watch) stops queueing others after 10s. Queued maintenance operations run one after another, and a waiting command waits for all of them. The queue marker is an OS lock beside the gate (*.gate.lock.intent, covered by the existing *.gate.lock* gitignore pattern), so a crashed waiter never leaves a stale queue behind. bd init's own wait (BEADS_INIT_GATE_TIMEOUT) rises from 30s to 60s, since back-to-back inits under load take 10-15s each.

  • Concurrent bd init --shared-server runs in different projects no longer refuse each other. Every shared-server project gates the one shared dolt data dir, and bd init holds that gate exclusively for its ~8s run but waited only 5s for it, so a second init (or an init during another project's long command) failed with "bd init refuses to run over live bd activity". bd init now waits (60s by default, see the gate-wait entry above), printing one "waiting for another bd process on the shared server" notice after 2s, then fails with the same refusal naming the bound. Override it with BEADS_INIT_GATE_TIMEOUT (2m, 90). Other exclusive operations keep their 5s wait.

  • A proxied-server command against an unreachable external Dolt upstream now fails within about a second with a clear error instead of stalling ~20-30s. The local db proxy stayed up, so the client only saw a bare close, which the bootstrap ping retried as a transient drop for its whole 30s budget before reporting invalid connection. The proxy now answers such a connection with a MySQL error (2003, prefixed beads db proxy:) saying what happened: the dial was refused / the socket is missing / the host is unreachable, or the upstream closed the connection before the MySQL greeting (down, restarting, or at its connection limit). The client retries that error only for about a second, so an endpoint that is flapping or rebinding is still ridden out within one command. A backend that drops a connection after the greeting, a dial timeout, and every refusal from a managed (local sidecar) backend keep the full transient retry.

  • bd ready --explain no longer reports a pinned dependency as a resolved blocker. The ready query skips a pinned target exactly as it skips a closed one, so a blocks edge onto a pinned bead never fences its dependent — but the explanation listed every blocking edge of a ready issue under Resolved blockers, status unread, and a bead wired behind a long-lived pinned bring-up reported itself satisfied while its precondition was unmet. Each blocking edge is now sorted by the target's status: closed under Resolved blockers (and resolved_blockers), pinned under Pinned dependencies (never block) (pinned_dependencies), and any other status under Open dependencies (not blocking) (open_dependencies) — the shape #6066 reports for a foreign-prefix blocker, now visible instead of passing as resolved. Reason keeps its N blocker(s) resolved lead and appends the pinned and open counts as clauses. Both routes fetch the ready issues' dependency targets alongside the blockers they already fetched. Pinned-never-blocks itself is unchanged.

  • bd list no longer silently drops all but the last repeated filter flag. --status, --state, and --id were plain string flags, so bd list --status open --status closed --status pinned kept only pinned — a census over 477 issues quietly answered over 3, with nothing in the output to distinguish the narrowed answer from a correct one. Repeats of those three flags now accumulate (duplicates collapse) instead of overwriting, so --status open --status closed selects the same set as --status open,closed. The one repeat that does not simply accumulate is one involving --status all: all cannot be combined with other statuses, so --status all --status open is now refused rather than narrowed to open. --type and --assignee are single-valued all the way down — unioning would fail type validation or exact-match nobody — so a repeat of either now refuses loudly instead of silently keeping the last value: invalid argument "epic" for "-t, --type" flag: --type given more than once (already "bug"); pass a single value. Single-flag and comma-form spellings behave exactly as before.

  • bd restore, bd admin compact, bd repo <add|remove|list|sync>, bd migrate [sync|hooks|schema] and bd preflight no longer register a local --json that shadows the root persistent flag. pflag keeps a command's own flag and drops the inherited one, so rootCmd.PersistentFlags().Changed("json") stays false for that command; #6293 compensated for that per command by making the pre-run consult the subcommand's own Changed bit before the root's. Deleting the eleven local registrations removes the shadowing itself rather than working around it: --json moves to the Global Flags section of those commands' --help, and a guard test now fails the build if a command shadows the root flag again — closing the class before an unbound local copy can reappear and invert the output mode it was meant to select.

  • bd init --force can no longer silently recreate a missing server-side database as empty (be-5up5). --force is an alias for --reinit-local, which skips the existing-data guard entirely, and the reinit path's own typed confirmation keys on the issue count — which is 0 in exactly the case where the database is missing. So bd init --force against a lost database created a fresh empty one at the same name, with no prompt and no destroy token: the 2026-08-11 fleet-wide data-loss reflex, reached through the flag an operator in a panic is most likely to try. The missing-database refusal now applies on the reinit path too. --force/--reinit-local authorize destroying a database that exists; only the explicit, per-invocation --recreate-missing authorizes creating one where the configured database has gone missing.

  • bd init --reinit-local and --force no longer wave through a workspace that predates project_id. Their missing-database safety net returned early whenever .beads/metadata.json carried no project_id, before it ever probed for this project's local database directory — so a workspace initialized before GH#2372 minted that field, whose local Dolt storage had since been lost, was protected under plain bd init but not under the two flags an operator reaches for in a panic. The probe now runs first and unconditionally, and an absent project_id is allowed through only when there is no local Dolt data directory either, matching the plain-bd init guard. --recreate-missing still opens it. The signal decision behind both guards is recorded in engdocs/adr/0004-missing-database-guard-signal.md.

  • bd init --recreate-missing now reaches its own opt-in. The existing-data check derived "is this project's database here?" from the Dolt data directory, which exists on every initialized workspace, so the missing-database branch was unreachable and both --recreate-missing opt-ins behind it were dead code. Four surfaces — the flag help, bd help init-safety, docs/recovery/init-safety.md, and the guard's own refusal text — told the operator to run bd init --recreate-missing --prefix <p>; that command exited 1 with a false "already initialized" message and pointed back at bd init --reinit-local, which is the destructive path, leaving the documented recovery circling between two refusals. The check now probes <dolt-data-dir>/<database>/.dolt, the same probe the reinit-path guard uses.

  • The missing-database guard probes this project's own database directory (<dolt-data-dir>/<database>/.dolt) rather than the Dolt data directory that merely contains it. The data directory always exists on an initialized workspace — shared-server mode creates the machine-global one on resolution, and per-project mode creates it at init — so probing it made the guard a no-op in shared-server mode, which is the topology of the 2026-08-11 loss. User-visible wherever the data directory exists but the database is gone.

  • The missing-database refusal no longer prints a ready-to-paste bd init --recreate-missing command line. It names the flag and points to bd help init-safety, which keeps the full invocation behind the diagnosis it requires (ADR 0002 Invariant 4). The refusal also fires when the server merely could not be reached, and a stopped server, a wrong port or a wrong data dir all look the same from there; recreating in any of them strands the real data behind an empty namesake. Run without --prefix, the old line also printed --prefix with no value. The --force and --reinit-local help, bd help init-safety and the recovery playbook now say that neither flag authorizes recreating a missing server-mode database.

  • BEADS_DOLT_POOL_READ_TIMEOUT / dolt.pool-read-timeout (and the write twins) now apply to every bd command in server mode. The knobs shipped in #5089, but their env/config ladder ran only for callers of NewFromConfig*; the CLI's own store open hand-builds its config and goes straight to New, so every bd command kept the built-in 10 s pool deadline whatever the knob said — on a large shared server that is what made bd close of a bead with dependents die in its recompute with i/o timeout and no relief valve. The ladder now runs from the constructor every DoltStore open shares — the CLI's store and library callers of New/NewFromConfig*; bd serve's HTTP data path builds its own DSN without pool deadlines and is unchanged (#6144). Note for operators of loaded servers: the documented precedence now reaches bd import as well — it used to inherit the 5 m long-read fallback unconditionally because the CLI's knob value was always 0, so a BEADS_DOLT_POOL_READ_TIMEOUT set below that now bounds import too; size the knob for your largest import, or leave it unset for the fallback.

  • BEADS_DOLT_MAX_CONNS / dolt.max-conns now size the pool for ordinary bd commands in server mode. The pool-size knob (GH#3140) had the same hole as the deadline knobs above: it was read only on the NewFromConfig* path (routed stores, bd doctor, ...), so the CLI's own store open always ran on the built-in 10-connection pool whatever the knob said. It now shares the deadlines' ladder in the constructor every DoltStore open reaches (#7052).

  • No-DB commands in a redirected workspace no longer lose the source repo's dolt_database (be-xil, be-fyt). bd doctor, bootstrap, context, dolt, init and version skip store init, and the beads dir they select is resolved through beads.FindBeadsDir(), which follows a .beads/redirect internally. The source repo's configured database name was therefore gone before anything could capture it, so bd doctor fell through to the shared target directory's own default database and reported Dolt Schema: wrong database — offering a --fix that would have repointed the rig at an unrelated rig's store. The redirect source is now captured before that resolution, matching the store-requiring path.

  • An explicit --db, BEADS_DB or BD_DB target is no longer shadowed by the ambient repo's redirect-source database (be-fyt). beads.GetRedirectInfo() deliberately resolves from the CWD repo's local .beads regardless of --db/BEADS_DIR (bd-wayc3), so when the ambient repo itself had an active redirect, the preservation above re-opened the same "wrong database" failure through a narrower door. Preservation is now skipped whenever the caller named an explicit target, asking the same question selectedNoDBBeadsDir asks so the two cannot drift apart. Note that a --db value naming a database rather than a path is not an explicit directory target: it is consumed only on the store-requiring path, the ambient workspace is still what gets selected, and its source database is still preserved.

  • An ambient BEADS_DOLT_SERVER_PORT now marks a workspace externally managed — suppressing auto-start — and no longer stops bd reaping its own orphaned server (#5934). Setting BEADS_DOLT_SERVER_PORT (or the legacy BEADS_DOLT_PORT) now makes ResolveServerMode classify a workspace as an externally-managed server, so bd stops trying to own a lifecycle it does not own. The stale-server cleanup path is deliberately carved out of that rule: it keeps resolving the mode without the port var, so bd still reaps a same-repo orphan it started (GH#2430) instead of declining because the environment named a port. A proxied-server workspace is exempt from the new rule entirely — it reaches its server through the proxy, so an ambient port does not describe its lifecycle.

  • bd caps how large the auto-backup destination can grow, instead of growing it forever (#6071). CALL DOLT_BACKUP('sync', ...) only ever adds new chunks to the destination — it never prunes ones that became unreachable on the source (history rewrites, superseded data) — and Dolt exposes no supported way to GC a backup destination in place. Left uncapped, the destination could only grow until disk filled; this is the root cause of the 2026-06-19 outage, where a 1.7GB store produced a 43GB backup directory. Auto-backup now pauses (nothing is deleted) once the destination reaches backup.size-cap-mb (default 2048MB); set it to 0 to disable the cap entirely. The pause is no longer stderr-only: bd backup status and its --json output now report a size_cap object (enabled, cap_mb, current_bytes, exceeded), so an agent/CI caller relying on --json or --quiet can see that auto-backup has stopped instead of reading a reassuring "Last backup" line while nothing further syncs. The remediation advice no longer suggests deleting the backup directory — nothing guarantees a deleted destination is cleanly recreated by the next sync, and the server-side backup remote stays registered against that path; it now points at raising backup.size-cap-mb or pointing backup.git-repo at a different git repository instead — not bd backup init, which configures manual bd backup sync's destination and leaves auto-backup paused. The size-cap check itself runs after both the interval throttle and change detection, so it costs nothing on any path that is not about to sync — an idle workspace never reaches it at all, and a paused destination re-arms the interval throttle on the skip, so it is measured at most once per backup.interval rather than on every command. backup.size-warn-interval (default 24h) controls how often the pause is re-announced. Manual bd backup / bd backup sync are not capped.

  • bd context resolves a workspace outside a git repository instead of hard-failing (#4772). The command's own help promises it "does not require the database to be open, making it useful for diagnostics in degraded states", but it exited with cannot determine repository root whenever the working directory was not inside a git repo — even with a perfectly valid .beads/ present — so consumers using bd context --json as a health probe got a false "unreachable" in healthy non-git scopes.

    GetRepoContext() finds and boundary-validates .beads/ before it asks git for a root, so that one failure means only that git is missing. It now returns a typed *beads.NoRepoRootError, and GetRepoContextAllowingNoGit() selects it with errors.As and roots the context at the .beads parent. Typed rather than by message text on purpose: the SEC-003 unsafe-location rejection embeds the offending path verbatim, so a substring test would let a workspace under a hostile path clear its own security refusal. Every other failure, that one included, propagates untouched.

    Scoped to the callers that only read config: bd context on both routes, and the domain/fs context provider — which bd serve also resolves through, so bd serve in a non-git directory with a valid .beads/ now proceeds rather than failing at context resolution. Anything that runs git commands still uses GetRepoContext(). The synthesized context roots at the repository containing the .beads/ (falling back to its parent when git cannot be asked there either) and decides is_redirected positionally — whether discovery standing in the working directory would have found that .beads/ — so --db, BEADS_DB/BD_DB and -C are reported the same way BEADS_DIR is. bd context also prints cwd repo: git: unavailable rather than silently omitting the one degraded state it exists to report.

  • Server-mode bd federation sync and bd federation status work with Dolt remotes that have no stored peer credentials (#5128, fixes #4837). A remote added by bd federation add-peer without --user or by bd dolt remote add has no federation_peers row, and every server-mode bd federation push, pull and fetch to it failed at the credential lookup with failed to get peer credentials: not found: federation peer <name>. A missing row now means no stored credentials, as it already did in embedded mode, so the operation runs with Dolt's own credential discovery: bd federation sync without --peer syncs those remotes too, and status fetches them. Errors reading or decrypting a stored row still fail. A mistyped --peer now fails with Dolt's own unknown-remote error instead.

  • bd comment/bd comments add refuse an abbreviated id and a reserved word instead of silently resolving one to the wrong issue (#5393). bd comment list <id>, a typo for bd comments list, used to parse as id="list" with no guard on the id slot — and since no issue is ever literally named "list", the default abbreviation-tolerant resolver would fuzzy-match it against whatever existing bead or wisp id happened to start with "list" and write the rest of the command line there as a comment. Confirmed in production: 15+ automated sessions made this exact typo over two days and all landed on the same unrelated wisp. bd comment's id positional now (a) refuses the four words a bd comments <verb> typo is likely to produce — list, add, rm, delete — before any id resolution is attempted, on both the embedded and proxied-server paths, and (b) requires an EXACT id match (bd-a1b2c3d4, not a1b2) rather than the abbreviation-tolerant matching every other command keeps. The refusal message is truthful about which case applies: a reserved word gets a hint toward bd comments; an abbreviation of a REAL issue gets id abbreviations are not accepted on comment writes; use the full id from \bd show `` rather than the generic (and in that case false) "no issue found matching" text a genuine not-found gets. See Working with IDs for the full read/write distinction. The proxied-server comment path was investigated separately and found to already be exact-id-only by construction (its resolution never performed abbreviation matching in the first place) — a regression test now locks that in rather than leaving it undocumented.

  • bd delete no longer resolves an id it cannot find to a different issue whose id begins with it. (#7375). The direct (embedded and server) route resolved each id with the abbreviation-tolerant resolver, so an id that was already gone (a retried delete, a re-run --from-file list, a parent whose child survives) resolved by leading-prefix match to whichever surviving issue's id it begins: after fx-t00003 was deleted, bd delete fx-t00003 previewed fx-t000031 and printed bd delete fx-t000031 --force as the next step, and with --force deleted it. --quiet silenced the partial-id notice, so a --force --quiet re-run deleted issues it never named without a word. bd delete now requires an exact id on the preview and with --force, as bd comment does and as the team-server route already did; full ids, bare full hashes, child ids and routed ids resolve as before. An abbreviation is refused with issue "t000" not found: bd delete needs the full issue id, and "t000" is only the start of fx-t000031, naming every id it matches; under --if-revision it is the guard's lost race instead (precondition_failed, exit 13), as for any other gone id. This is a user-visible change: a script or habit that deletes by abbreviation now gets that error and must pass the full id. Reads and other writes keep accepting abbreviations; see Working with IDs.

  • A git hook beads never wrote is no longer reported as installed (#6084). getHookVersion returns no error for a readable hook file that carries no beads markers, so a hand-written pre-commit — one beads does not own — was printed as ✓ installed (version ) by bd hooks list, counted as installed by bd info, and reported as no drift by bd config drift. Those surfaces now report such a file as not installed. This is a user-visible change to a scripted contract: in a repository whose hooks beads does not own, bd config drift now exits 1 (drift detected) where it exited 0, so automation keyed on that exit code should re-check those repositories before upgrading. A hook that calls bd hooks run <hook> still counts as installed even with no beads marker — that is the integration beads prescribes for external hook managers such as lefthook and husky (GH#946) — and it is version-agnostic, so it is never reported outdated and never prompts bd hooks install, which would overwrite the manager's own hook. bd hooks list, bd info and bd config drift now use the same predicate as bd doctor for deciding whether a hook file is a beads hook; bd doctor's own installed/missing check still only tests for file existence and is tracked separately.

  • bd close verifies a bead gate in proxied-server mode. The close pre-check built its bead-gate lookup on the local store, which the proxied-server route never opens, so closing a bead gate there refused with no local store available even after the awaited bead had closed, while bd gate check resolved the same gate. The close pre-check now reads the awaited bead through the same fresh-read path bd gate check uses on that route; the direct and embedded routes are unchanged (#5861 item 1).

  • bd recall, bd forget, bd remember <bare-key>, GET /v0/beads/memories/{key} and DELETE /v0/beads/memories/{key} now treat an empty-valued memory as present (#5964, fixes #5963). Presence used to be derived from the stored value's truthiness, so a memory written out of band with an empty string was invisible even though its row existed. bd recall printed "No memory with key" and exited 1, bd forget refused to delete it, bd remember <bare-key> refused to recall it, and both HTTP doors answered 404 (the DELETE leaving the row in place), while GET /v0/beads/memories already enumerated the same row. Presence now comes from row existence on direct Dolt, embedded Dolt and the proxied backend. An empty-valued memory is found, recalls as an empty string, and is deleted on the first forget. Only a key with no row is a miss.

  • bd doctor no longer flags a .local_version that starts with v. The canonical spelling of a Go module version — and the string a build stamped from a Go pseudo-version reports and writes into .local_version itself — was rejected by the Version Tracking check (Invalid version format), and read as major 0 in version comparisons, because the helpers parsed digits first. They now accept an optional leading v; versions without it are unchanged (#6152).

  • bd serve's blocked-close refusal names the blockers. A 409 not_closable for a live blocker — on issues/{id}:close, issues:batchClose per-item outcomes, PATCH issues/{id} into a done status, and issues:batchApply — said only "issue is blocked", so an HTTP client could not tell the user what held the close, while bd close on the direct and proxied routes prints cannot close blocked issue: X is blocked by [Y]. The refusal now carries a blockers extension member ([{id, kind, type}], kind local or external, type the blocking edge type when known) and a detail that opens with the direct route's sentence. The code, the status and the open_children discriminator are unchanged. The list travels typed end to end: close refusals are now an issueops.BlockedError (still matching ErrCloseBlocked, with a byte-identical message), so the server never parses prose to build the member and a client can rebuild the same typed error.

  • The #6716 fan-in stall is fixed on the proxied-server route and the remaining store routes. Two blockers of one dependent taken away at the same time (parallel workers closing both, or a close racing a bd dep remove or a delete of the other) could leave the dependent is_blocked=1 and missing from bd ready until bd recompute-blocked (#6716). The post-commit recheck the Dolt store write transactions gained for that issue now also runs on every write the proxied-server (uow/domain-db) route serves, which under --proxied-server — the default topology — is bd close (single and batch), bd update --status, bd dep remove, bd delete, bd batch and bd serve; on RunInTransaction (bd batch direct, bd cook, bd mol squash/burn, and SDK callers that close or update inside a transaction); and on the wisp close, update and delete writers and the legacy dependency removal behind bd duplicates --merge. Demote-to-wisp moves a row between planes without taking a blocker away and is unchanged.

  • An explicit BEADS_DIR is authoritative during workspace discovery. When BEADS_DIR named a directory without project files yet (missing, empty, or not initialized), discovery ignored it and walked up from the current directory instead, then rebound BEADS_DIR to whatever ancestor workspace it found. bd init refused with "already initialized" because of the parent workspace, and data commands such as bd create and bd list read and wrote the parent's store. Discovery now reports no workspace for such a BEADS_DIR, so bd init initializes the named directory and other commands fail with "no beads database found". bd import/bd setup and bd bootstrap target the named directory when they create a workspace. Behaviour with BEADS_DIR unset is unchanged. Behaviour change: a BEADS_DIR that points at a project root rather than its .beads directory (for example BEADS_DIR=$repo instead of BEADS_DIR=$repo/.beads), or at a .beads that does not exist yet, used to work by accident because discovery walked up to the nearest workspace; it now fails with "no beads database found". Point BEADS_DIR at the .beads directory itself. (#6938)

  • A proxied workspace's proxy retires when its Dolt backend exits cleanly. The proxy noticed its dolt sql-server child exiting only when the exit status was non-zero. A backend that shut down gracefully (for example on SIGTERM) left the proxy up and adoptable in front of a dead store: with --proxied-server-idle-timeout 0 it never went away, bd dolt status called it running, and every command failed until bd dolt stop. The proxy now retires on any backend exit, as it already did on a crash, so the next bd command (including bd ping) starts a fresh proxy and backend. bd dolt status no longer reports a proxy whose managed backend is gone as running: the text says "not serving" and the JSON reports running: false (with proxy_pid still set), including for a stranded proxy started by an older bd. (#6937)

  • bd sql no longer drops the rows of CTE queries or CALL result sets (#6932). In proxied-server mode, WITH name(cols) AS (...) SELECT ... and WITH RECURSIVE ... queries were misread as writes and printed OK, 0 rows affected with no error; they now return their rows. Statements are now classified with the Dolt SQL parser (shared by the proxied and direct paths): reads return rows, plain writes (including CTE-prefixed UPDATE/DELETE) commit and report rows_affected, and anything that may both write and return rows (CALL, EXPLAIN ANALYZE of a write, RETURNING) or that the parser cannot classify is committed and prints whatever rows it returns, so CALL DOLT_BRANCH(...)-style result sets are rendered instead of discarded. Visible changes for scripts: in direct server mode a multi-statement write, or a single write statement the parser cannot classify, now prints OK / {"status":"ok"} (as proxied mode already did) instead of OK, N rows affected / {"rows_affected":N}; and --readonly now refuses statements it cannot parse (for example PRAGMA) instead of treating them as reads.

  • bd types now lists exactly the types bd create --type accepts (#6934). The two resolved the type set through different code and could disagree: a server-mode workspace's bd types dropped .beads/config.yaml types.custom entries once the database had any custom type, although create accepts them; in proxied mode bd config set storage-class.<type> checked config.yaml alone and refused custom types registered in the database; and the built-in system types message, molecule, gate and event were never listed. Every mode now composes custom types by one rule (the custom_types table, else the types.custom config row, always unioned with config.yaml). bd types gains a "System types" section, and bd types --json an additive system_types field; core_types and custom_types are unchanged. A failure to read the custom types now fails bd types, bd create --graph and bd config set storage-class.* instead of being silently treated as "no custom types".

  • bd refuses a bare tooling sentinel (null, undefined, an empty string) as an issue ID (#6214). jq -r prints the literal null when a selector misses and JS tooling prints undefined, so bd update "$(bd list --json | jq -r '.[0].id')" --status closed arrived at ID resolution as a plain four-character token. Those tokens are a valid partial-ID shape, so they reached the leading-prefix abbreviation branch and resolved to whichever issue's hash began with them — closing an issue the caller never named and exiting 0 with a success line. bd update null, bd comment null and every other command that resolves an ID now fail before any lookup, because the guard sits in the shared resolver. An ID that merely contains a token (bd-null, null3t0, undefined-behavior) still resolves, and abbreviation matching is otherwise unchanged. The empty string already failed; it now says why.

  • Table-rebuild __temp__ intermediates can no longer materialize as tracked tables on @@dolt_transaction_commit=1 servers. The ignored-series rebuilds stage every clone-local table through a __temp__<table> rename, but the intermediate names carried no dolt_ignore pattern: against a transaction-commit server each CREATE TABLE __temp__X auto-committed a real table at HEAD, and the rename onto the ignored final name left an unstageable __temp__X -> X half in dolt_status that wedged every later open behind the dirty-table guard (observed in the field as a three-city shared hub wedging twice in one day, each time ending in fleet-wide write refusal and a manual SQL repair). __temp__% now ships in the canonical seeded pattern set — asserted, like the events-journal patterns, before the series that creates the tables — so intermediates are born ignored. The scoped seed commit also gains --skip-empty: under @@dolt_transaction_commit=1 the server has already committed the seeded rows at their own transaction boundaries, and the previously unconditional DOLT_COMMIT died with "nothing to commit", killing the pass on exactly the under-seeded stores the heal targets. Fenced deployments: adding a canonical pattern un-converges dolt_ignore for every existing database, so the next write-mode open issues one INSERT IGNORE INTO dolt_ignore; a hosted or box-fenced wire client that is denied that INSERT fails its open until one privileged bd opens the store and heals the pattern — self-healing, one privileged open per upgrade (degrading a denied seed to a skip is tracked separately).

  • bd ready --parent and bd blocked --parent no longer re-scan the whole parent-child edge relation for every descendant they find. The transitive descendant walk recursed against a materialized parent_edges CTE that Dolt cannot index through, so its cost was (parent-child rows) × (descendants): 7.5 s for a 483-descendant parent on a 4.6k-edge database, and past the shared-pool read deadline on a busy server. The walk now recurses directly off dependencies / wisp_dependencies through their typed target indexes and returns the same rows in a fraction of the time (#6128).

  • bd list --parent on a proxied-server workspace no longer drops a child whose prefix differs from its parent's (#6130). The edge to such a parent is stored in depends_on_external — issueops.IsExternalDepTarget routes every cross-prefix target there — but the descendant walk behind the proxied tree view resolved a parent only through the issue and wisp target columns, so the child and its subtree were missing from the tree and from --watch, while the direct route listed them. The walk now reads all three target columns, as the direct route's --parent filter does. Nothing listed before can drop out: the added column is read only when the other two are both empty.

  • notion.token is kept out of the Dolt database (#6676). It was missing from the yaml-only key list that holds the other tracker secrets, so bd config set notion.token — the command bd notion status recommends — wrote it to the database, whose contents bd dolt push sends to the remote. It is now written to config.yaml like github.token, gitlab.token and the other tracker secrets, and refused on a git-tracked config.yaml with a pointer to NOTION_TOKEN. A token an older bd already stored in the database is still read, after config.yaml and before NOTION_TOKEN, so existing setups keep authenticating. bd config unset notion.token removes both copies — the config.yaml entry and any row an older bd left in the database — and reports which of them it actually removed; that cleanup covers every yaml-only secret key, not just this one. The database half is best effort and says so: it is skipped only where reaching the row would mean creating a database (the command never creates one), so it still runs for a server-mode or proxied workspace whose database is simply not on local disk, and it is attempted even when the config.yaml edit fails, so a workspace whose only copy is the stored row still has a remover. bd notion status now reports the auth source as database_legacy rather than config_token while the stored row is the one authenticating, so an affected workspace can identify itself. A token that was already pushed should still be rotated: deleting the row locally does not unpublish it from remotes that already have a copy.

  • A server-mode workspace with an empty .beads/dolt is no longer refused as legacy (#5682, #6935). bd creates .beads/dolt on use even when the data lives on an external Dolt server, and leaves it empty. When the gitignored .local_version witness was missing — after a fresh checkout, or when a provisioner created the empty root before bd init --server --external — the legacy-upgrade guard refused every command, bd init included, as a "legacy Dolt server workspace", and no bd command could repair it. An empty root holds nothing a legacy release could have left, so the guard now admits it: bd init writes the witness, and an existing workspace in that state re-seeds it on the next command. A pre-1.0 witness, or a .beads/dolt that holds anything, is still refused.

  • bd -C dir prime now describes the target workspace instead of the launch directory (#5509). -C resolves BEADS_DIR but never changes directory, so prime's cwd-relative surfaces still described wherever the command was launched: the clone-local .beads/PRIME.md tier, the AGENTS.md/CLAUDE.md divergence reminder, the git upstream/remote probes behind the template's git-authority wording, and the redirect notice. All four now resolve against the -C target, matching cd dir && bd prime; without -C the behavior is byte-for-byte unchanged.

    Note for -C users: the git-remote probe prime shares with auto-backup (isBackupAutoEnabled) follows the same change, so under bd -C dir <any command> auto-backup enablement and the bd backup status note are now decided by the -C target's git remote rather than the launch directory's. That makes the probe agree with the store actually being backed up, but it can flip auto-backup on or off for -C invocations whose launch directory and target differ in remote configuration.

  • bd show --watch works under --proxied-server. It was refused with proxy.watch.unsupported although the provider can answer the poll the way bd list --watch already does, so every proxied workspace — the default transport for managed-local setups — had no way to watch a bead. The proxied route now shares the direct route's loop: render once, re-read every 2s, redraw only when the issue's status or updated_at changes, stop cleanly on Ctrl+C/SIGTERM, and require exactly one id. Each poll opens its own short unit of work, so a long watch never pins a transaction, and a poll that fails (the issue was deleted, the backend blipped) keeps the last render without printing, as on the direct route. On both routes a watch whose id cannot be found now exits non-zero instead of 0.

  • Generated git hooks accept uutils coreutils timeout as a deadline helper (#5541). The managed hook section probes timeout and gtimeout with --version and accepted only the GNU coreutils banner, so a host whose timeout is uutils coreutils (Ubuntu 25.10+, or a distribution configured with uutils-coreutils in place of GNU) failed the probe and fell through. Where Perl was installed the shim still got a deadline from the Perl alarm arm — which Git for Windows Perl does not guarantee across exec — so the missing deadline bit uutils hosts without Perl, which ran bd hooks run unbounded (with the documented warning). The probe now also accepts the timeout (uutils coreutils) banner; native Windows timeout.exe stays rejected (#5503). The widened probe ships in the generated hook section, which is rewritten only by bd hooks install — hosts that already installed hooks must run it once to pick up the fix.

  • bd close now exits non-zero when any issue in a batch fails to close (#6648). A batch with one refused id used to exit 0 as long as another id closed, so scripts could not tell that part of the batch was left open. The closable ids still close, the refusals are still printed, and a final Error: N of M issues failed to close line is added; ids that were already closed still count as success. Both the direct and proxied-server routes behave this way for ids the close policy or the engine refuses. (An id that cannot be resolved at all is unchanged and still differs by route: the direct route aborts the whole command before any close runs, while the proxied route refuses that argument and closes the rest.) In --json mode the summary is instead a compact JSON line on stderr naming the failed ids, matching bd update's partial-failure report, while stdout keeps the usual closed-issues array. Each failed[] entry carries the refusal as the engine worded it, identically on both routes; the --force hint and the route's own framing stay on the human-readable stderr line, which is unchanged. --claim-next still claims when part of the batch closed — the claim commits inside the batch's own transaction and a sibling's refusal does not roll it back — so the summary names the claimed id rather than leaving it silently assigned.

  • Unblocking two blockers of one dependent at the same time — closing both, or a close racing a bd dep remove or a delete of the other — no longer leaves the dependent stuck as blocked and hidden from bd ready until bd recompute-blocked: a write that takes a blocker away (a close, an update to an inactive status, a dependency removal, a delete) and runs through a Dolt store write transaction now rechecks the dependents it recomputed once that transaction has committed (#6716). The proxied-server route and the remaining store routes are covered by the entry above.

  • bd list --watch --format is refused instead of silently dropping the format (#6277). --watch always re-renders the pretty listing, so on the direct route a --format template was ignored without a word, while --proxied-server already refused the combination. Both routes now fail with the same --format cannot be combined with --watch usage error.

  • Explicit bd update --claim and direct bd close now enforce unsatisfied external capability dependencies. Mixed close batches still commit eligible items together, and --claim-next skips externally blocked work even when --force overrides the close guard.

  • The smart migrate gate no longer auto-migrates a clone whose data is behind the remote, and bd dolt pull now works from that state (#6575). The gate's first-mover verdict was computed from schema facts alone, so a clone level with the remote on schema but behind it in unpulled data commits was classified a safe first-mover and migrated in place — minting local-only schema commits on a history missing those commits, which can leave every later bd dolt pull refusing to merge (#6368). The equal-version path now reads the branch position its callers already supply and stops with a pull-first remedy instead.

    The stop applies whenever the clone has commits left to pull, whether or not it also has local commits of its own. Since bd auto-commits every write, a clone that is both behind and ahead is the ordinary multi-machine state and reaches the same wedge; it is told its pull will merge rather than fast-forward, and how to resolve conflicts if it reports them. A clone that is level, or that only has unpushed local commits with nothing to pull, still auto-migrates exactly as before.

    Crucially, the refusal no longer blocks its own remedy. bd dolt pull opens the store too, so on a refused clone it used to hit the very refusal that prescribed it — and an embedded workspace has no external dolt binary to fall back to, which left BD_ALLOW_REMOTE_MIGRATE=1 (performing the migration the stop exists to prevent) as the only way forward. bd dolt pull now opens leniently for this one refusal, the way bd dolt commit does for the dirty-working-set refusal (#4566); every other gate refusal still fails that open. This supersedes the 1.3.0 upgrade note's statement that the pending-migration gate refuses bd dolt pull: for the data-behind stop it does not, by design. Re-run the command you were blocked on once the pull completes and the migration proceeds.

    Commands that keep working through the stop say the same thing. bd list, bd ready and bd show succeed against the old schema, and bd dolt commit still commits the working set; on embedded storage all of them used to print the blunt migrate-or-adopt coordination bullets while doing so — naming bd migrate --force && bd dolt push, which on a data-behind clone applies the migration this stop exists to prevent and then fails the push non-fast-forward, and bd bootstrap, which no-ops against an existing workspace. They now print the pull-first guidance, shape-branched the same way the fatal refusal is, alongside the note that the command in hand continued at the current schema. Refusals that are not the data-behind stop keep the coordination bullets, which are right for them.

    --json callers get the same remedy the terminal does: observed, expected and options describe the pull (a single pull-first option) rather than the migrate-or-adopt decision that does not apply here, plus a data_behind_shape of fast-forward or diverged. For the fast-forward shape on a non-shared store — one unconditional option, no local commits, nothing discarded — human_decision_required is now false, so an agent can run the pull instead of stalling for approval of a step the same payload calls riskless. It stays true for the diverged shape (the pull merges and can need conflict resolution) and on a shared store (the follow-up consent step needs an operator who can confirm every co-resident client is upgraded). On a shared Dolt sql-server the guidance still carries #5920's consequence — migrating promotes the schema for every co-resident client — and names the consent step the retry needs there: bd migrate schema --force. It is the forced form because this stop always has a remote configured (behind-ness is read from the remote-tracking ref), and the bare bd migrate schema consent is only read for a shared database with no remote; the flag consents to migrating a remote-backed shared store, and by then the pull has landed the commits the clone was missing. The payload's docs link points at the data-behind section rather than the migrate-or-adopt recipe, and that section — plus the upgrade and init-safety ordering rules — now records the bd dolt pull exception instead of stating the pull is always refused.

    Proxied-server mode reaches the stop too, and is told where to run the remedy. The store-open gate on that path supplied no branch-position callback, so a proxied clone that was level on schema and behind in data could never be classified data-behind: it got the blunt shared-store refusal, whose body is the designated-migrator recipe — bd migrate --force then bd dolt push — which in this state is the wedge the stop exists to prevent. It now routes to the same data-behind stop as every other topology. Because bd dolt pull is refused at the proxied front door (proxy.dolt_pull.unsupported), along with bd dolt push, the bare bd migrate and bd conflicts, the guidance there names the machine the pull has to run on rather than printing a command this binary rejects; the --json option is pull-first-on-server-host, and expected carries the same qualifier so a single-field reader is not handed a locally-refused command. The shared-store consent step is the other way round: bd migrate schema --force is not refused here — the capability registry permits the two-word migrate schema path — so the guidance says it can be run from this workspace but only after the pull lands on the server host, and the warning against forcing past the stop names it alongside BD_ALLOW_REMOTE_MIGRATE=1 as the two consent surfaces this topology can still reach. Read-only proxied opens print the same pull-first block instead of the shared-consent template. The Clone behind the remote docs section the payload links to, and the upgrade and init-safety ordering rules that point at it, now give the same proxied-server answer.

    Both existing escape hatches are unchanged: bd migrate --force / BD_ALLOW_REMOTE_MIGRATE=1 are still consulted before the smart gate, and BD_SMART_GATE=0 still yields the blunt gate unconditionally (including for bd dolt pull, which is what opting out means). Pulling before the first open after upgrading avoids the state entirely.

  • A --format typed on bd list now beats a json: true config default (#6278). With json: true in config.yaml (or the environment), bd list --format digraph, dot or a template printed the JSON listing instead, because bd list's own --format flag shadows the root one the config-precedence check looks at. A typed --format now wins over the config default; an explicit --json still wins over --format.

  • Strict readonly no longer records last-touched state (#5785). Under strict readonly (--readonly, or readonly in config), bd show wrote the issue id to .beads/last-touched even though readonly promises not to mutate workspace-local state — the database store was opened read-only while this marker file was still written. SetLastTouchedID now returns early in that mode, so a readonly workspace leaves the marker untouched. Ordinary (non-readonly) bd show keeps recording last-touched exactly as before, so only consumers driving bd --readonly … --json see the change.

  • Markdown bodies now honor the terminal's actual background instead of always rendering dark (#5776). glamour.WithEnvironmentConfig() defaults to styles.DarkStyle whenever GLAMOUR_STYLE is unset, with no regard for the terminal's real background — unlike internal/ui's own adaptive colors, which probe it via lipgloss.HasDarkBackground. That left every markdown body's colors wrong on a light-background terminal even though the section header wrapping it adapted correctly. The fix lands at the shared renderer, so it covers every markdown body bd prints — DESCRIPTION, DESIGN, NOTES, ACCEPTANCE CRITERIA, and comment bodies, direct and proxied alike — not just bd show's DESCRIPTION field. An explicit GLAMOUR_STYLE override still takes priority over the detected background.

Added

  • POST /v0/beads/issues:batchApply accepts up to 1000 items, up from 100. issueops.MaxApplyBatchItems moves from 100 to 1000; the request body cap rises from 4 MiB to 16 MiB (maxApplyBatchBodyBytes); and the whole plan still applies in exactly ONE transaction — the cap raise never chunks a request. A request over 100 items now runs under a separate, EXTENDED whole-run budget instead of the ordinary 60s request deadline: a new bd serve --large-apply-ceiling flag sets it (5 minutes by default), applied flat regardless of item count, never scaled down. A request at or under 100 items is completely unaffected and keeps exactly the 60s deadline it has always had. A request over 100 items also serializes against every other such request through a dedicated one-wide semaphore, so at most one oversized write transaction is in flight at a time; the wait for that semaphore is itself bounded, so a burst of oversized requests is shed with 503 rather than queuing indefinitely and starving ordinary traffic. An ordinary (<=100-item) request is never made to wait on it. Graceful shutdown waits out an in-flight large apply up to its own deadline, but refuses any large apply still queued for the semaphore once shutdown begins, rather than extending the drain further. bd serve's orchestrator stop grace must be at least --large-apply-ceiling plus 5 seconds, or an external SIGKILL can still cut off an in-flight large apply the drain would otherwise have waited out; see engdocs/SERVE_RUNBOOK.md. A new issues.batchApplyLarge capability token on GET /v0/beads/context lets a client learn whether a given server accepts the raised envelope before sending a plan over the old 100-item bound. Measured wall-clock for the whole-transaction apply (embedded Dolt, build+commit only) backs the 5-minute default: a 356-item plan (the design doc's primary measured shape) commits in ~12.6s (23.8x headroom under the ceiling), and a 1000-item plan (the new cap) commits in ~57.5s (5.2x headroom).

  • A long-running schema migration now says so instead of going quiet (#5997). Migrations are allowed to take a long time by design — migration 0047's full-table is_blocked recompute is a real example — but until now a slow one and a wedged one looked identical from outside: bd simply stopped printing. A watchdog now emits a WARN naming the migration's version, name, and elapsed time once the migration's own SQL has been running past the interval, and repeats every interval while it keeps running, so an operator reading logs can tell "still working" from "stopped emitting anything". Coverage is the migration body specifically — the per-step Dolt commit that follows it on the production embedded path is outside the watchdog, so a stall there is still quiet. Set BEADS_MIGRATION_WATCHDOG_INTERVAL to change the 5-minute default; it accepts durations like 10m and bare seconds like 90, and falls back to the default when unset or unparsable. This is observability only and never a circuit breaker: the migration receives the caller's exact context, its error is passed through unchanged, and nothing is aborted or rolled back. The warning is deliberately not terminal-gated, so it survives bd serve, systemd, CI, and piped invocations.

  • Auto-backup runs on a managed-local proxied-server workspace. The proxied arm of the post-command hook now calls auto-backup, so an explicit backup.enabled=true (or BD_BACKUP_ENABLED=1) takes the same throttled, change-detected Dolt-native backup into .beads/backup it takes in direct mode, through the proxied provider's non-transactional seam. Before this the opt-in was inert on every proxied workspace and bd backup status said so. The default stays OFF, as on every server-mode workspace: many bd clients share one server. It is honored exactly where bd backup sync is — a proxied server bd started itself — and stays inert on an external or team-server topology, strict --readonly, bd serve, a preview (--dry-run/--inspect) and a migration freeze. A command that opened no provider (e.g. bd dolt stop) backs nothing up rather than relaunching the server to do so. bd backup status now reports the proxied default as auto: off in proxied-server mode; set backup.enabled=true to opt in.

  • Auto-backup and bd backup sync no longer overlap. A per-workspace backup lock (.beads/backup.lock) serializes them in every mode: auto-backup skips when another backup holds it, and bd backup sync waits up to 5 seconds (the bound restore uses for the workspace gate) and then fails with "another backup is running for this workspace".

  • bd purge covers an orchestrator's wisp retention sweep. Three changes that together let bd purge --wisps-plane --older-than 168h --force replace a raw DELETE FROM wisps retention step:

    • Live-dependent protection, always on. bd purge no longer deletes a closed bead that a live bead depends on through a parent-child, tracks or blocks edge — a closed molecule root whose step is still open, a closed wisp a live convoy tracks, a closed blocker of live work. "Live" is any status that is not done, including custom statuses. The held-back count is reported as live_dependent_skipped in --json (and on its own line in text output). The protection is on the role (issueops.SweepRequest.ProtectLiveDependents); bd prune does not ask for it.
    • --wisps-plane selects every closed row stored in the wisps table, including --no-history beads, which the default ephemeral selection leaves to bd prune. It works on embedded, server and proxied workspaces, and because it reaches durable-tier rows it requires --older-than or --pattern, like bd prune (issueops.SweepWispsPlane).
    • --older-than accepts hour and finer durations (36h, 168h, 90m, down to 1s) on both bd purge and bd prune. Day values (7, 7d, 2w) are unchanged. An Nh value used to be converted to whole days: floored above a day (36h swept rows only 24 hours old) and rounded UP to one day below it (12h kept everything younger than 24 hours). Both are now taken exactly, so bd prune --older-than 12h now deletes rows closed 12–24 hours ago that it used to keep. A value too large to represent (e.g. 213504d) is now refused instead of silently wrapping to a tiny age.
    • --limit N caps one bd purge run at N beads, oldest-closed first (issueops.SweepRequest.Limit), so a large backlog drains in bounded transactions; --json then adds remaining and has_more. Loop while has_more is true.
  • bd backup works on a proxied-server workspace bd runs the Dolt server for. bd backup init, sync, remove, status and restore are routed over the proxied provider; before this, a proxied workspace — the default topology since 1.3.0 — had no backup path at all. bd backup restore stops the proxy and its Dolt child before replacing the database and leaves the workspace quiescent, so the next command relaunches against the restored data.

    The family stays refused by design on a proxied workspace pointed at a Dolt server bd does not own (an external host, socket, or a beads-team-server database): CALL DOLT_BACKUP('add', …) registers the backup remote on the server, where it is global to every client connected to it, so one workspace's backup decision would silently become everyone's — the shape bd backup's own help text already warns about for auto-backup. A file:// destination has a second problem on top, since the server resolves that path on its own filesystem. The refusal keeps the proxy.backup.unsupported code and now carries "reason": "design"; a backup story for those deployments belongs to whoever runs the server. Whether a remote-scheme destination (DoltHub, aws://, gs://) should eventually be allowed there is recorded as an open question rather than settled — the server pushes those over the network from its own environment, which is the same credential question the next slice has to answer for dolt push.

    Backups on this topology are explicit only. Auto-backup has no post-command hook on the proxied path, so backup.enabled=true does nothing there; bd backup status and bd backup help now say so instead of attributing the OFF to a missing git remote, which was never the reason on a server topology. Wiring an auto-backup hook through the new routes is left to a later slice.

  • bd says when it resolved a partial ID instead of matching one exactly. (#6302) ID resolution tries an exact match first and only then falls back to leading-prefix abbreviation (a3f8 -> a3f8e9). Ambiguity was already a hard error naming every candidate, but a sole abbreviation match was returned silently, so a caller could act on an issue it had not named without any signal. bd now writes a one-line note to stderr identifying both the input and what it resolved to. The case that motivates it: a hierarchical child is a valid leading-prefix match for its own parent's ID, so after a parent is renamed or deleted the vacated ID quietly resolves to the child. Note this covers every leading-prefix abbreviation, not only that shape, so the ordinary documented a3f8 -> bd-a3f8e9 path now emits a line too, once per resolved argument. That breadth is deliberate: a .-boundary rule would fix parent/child but leave bd-x.1 -> bd-x.11 just as silent. stdout and --json are unchanged; silence it with --quiet or BD_NO_PARTIAL_ID_NOTICE=1.

  • bd count accepts --has-metadata-key (#6160), closing the last metadata-filter gap against bd list, bd ready and bd search. A caller counting rows that merely carry a key no longer has to list them and measure the page.

  • bd count supports repeatable --metadata-field key=value filters (#6023), so callers can count the same metadata-scoped set bd list returns without fetching every row.

  • bd label rename <old> <new> fixes a label everywhere it appears, in one command instead of a per-issue add-then-remove loop (#6009). An issue that already carries the new label doesn't error — the rename drops its stale old label and counts the issue as a merge instead, reported honestly: Renamed label 'Backend' to 'backend': 6 issues (2 already had 'backend'). A label nothing carries is a clean no-op (No issues found with label 'Backend', exit 0). The sweep covers both planes, durable issues and wisps, and the direct and proxied-server routes end at the same issueops.RenameLabelInTx primitive — the proxied route journals through domain.LabelUseCase.RenameLabel rather than a per-issue add+remove fan-out, so merge semantics, the zero-carrier no-op, and the single label_renamed event per touched row are identical on both.

    --dry-run previews the blast radius (a count and the first 10 issues, all of them under --json) without writing anything, through a read-only path kept apart from the write path so a preview can never reach a transaction. Its merged count is a plain snapshot intersection of two independent reads, though, not the write path's measured one: the real rename derives merged from what its INSERT IGNORE actually affected, specifically so a concurrent label add between two reads can't skew the count, and the preview has no insert to measure against. Treat the dry-run number as an estimate — the flag's own help text says so now too. A rename that fails partway through reports an UPPER BOUND - up to how many issues may have been renamed, and up to how many of those were merges - rather than only the error. It cannot be a landed count: both routes assign those numbers inside the transaction, so a rollback or an exhausted retry returns them still describing an attempt that landed nothing. Only a Dolt publication failure after the SQL side already committed is a genuine partial, and the count alone cannot tell the two apart.

  • bd label remove --prefix removes every label matching a prefix in one call, instead of requiring an exact label name per removal. bd list/bd ready already supported --label-pattern/--label-regex for finding issues by label shape; label remove had no equivalent for stripping a whole label family (e.g. every pool:refused:* reason label) — callers had to fetch the labels themselves and issue one label remove per exact name. --prefix resolves the matching labels per issue and removes that issue's whole matching set in a single edit; bd label remove <id> --prefix pool:refused: replaces that fetch-then-loop. Routed through the same issueops.Lifecycle/ issueops.Reader roles as a plain label remove, so it works identically on the embedded and proxied-server (bd serve) storage modes.

  • PATCH /v0/beads/issues/{id} accepts claim: true, the wire spelling of bd update <id> --claim. It passes through to the same lifecycle role the direct route uses (issueops.UpdateRequest.Claim), so the claim and the patch land in one transaction with the direct route's semantics: a same-actor re-claim is idempotent (changed: false), a foreign holder is 409 already_claimed and an unclaimable status 409 not_claimable — both naming claim and carrying {id}:claim's assignee / issue_status members — and a refused claim writes none of the patch. With claim: true the patch may be empty. claim beside expected_assignee, expected_status or force_assignee_transfer is a 400. A server that predates the member refuses it as unknown_parameter before any database work, so an HTTP client can forward bd update --claim instead of refusing it, and can never mistake an older server for one that dropped the claim.

Fixed

  • bd show counts a wisp's comments instead of reporting comment_count: 0 (#5565). On the direct/embedded path, CountIssueComments always queried the permanent comments table, while GetIssueComments routes an active wisp to wisp_comments, so --include-comments returned rows the count said did not exist and comments_omitted was never set. The count now routes by wisp status the same way, matching proxied-server mode.

  • bd dolt start no longer puts a second sql-server over a proxied workspace's data directory. On a proxied-server workspace the default root IS .beads/dolt, and bd dolt start knew nothing about proxied mode: with the proxy quiesced it launched an unsupervised dolt sql-server directly over that root, after which ordinary bd commands failed (invalid connection) because the relaunched proxy found a foreign server on its port; with the proxy live it instead adopted the proxy's own dolt child into bd's classic PID/port records, leaving two managers for one process. It is now a typed refusal with code proxy.dolt_start.conflict. The proxy owns its backend's lifecycle — it starts on demand and bd dolt stop shuts it down. Direct-server and embedded workspaces are unaffected.

  • bd info honors doctor.suppress.git-hooks (#6027). bd info printed the "Git hooks not installed" / "outdated" warning unconditionally, so suppressing the check for bd doctor left it nagging on every bd info. It now reads the same doctor.suppress.* config through the doctor helper and skips the warning when git-hooks is suppressed. bd info --json never carried the warning and is unchanged.

  • bd dolt status tells the truth on a proxied workspace. It read the classic PID file, which proxied mode never writes, and so reported Dolt server: not running while the proxy was serving CRUD — the wrong answer that sent operators to bd dolt start in the first place. It now reads the proxy's own records and reports the proxy and its dolt backend separately, without starting either. JSON output shape change: on a proxied workspace bd dolt status --json now emits {"mode": "proxied-server", "root": ..., "running": <proxy up>, "proxy_pid": ..., "proxy_port": ..., "backend_managed": ..., "backend_running": ..., "backend_pid": ..., "backend_port": ..., "idle_timeout": ...} instead of the always-false {"running": false, "pid": 0, "port": 0}. running describes the proxy, the endpoint every bd command connects through; backend_managed is false on external proxied topologies, where the dolt server is not bd's process to report on.

  • bd create --metadata and bd update --metadata now require a JSON object (#6035). Both checked only that the value was valid JSON, so bd create --metadata '"oops"' stored a bare string (or an array or number) as the issue's metadata, while bd update refused the same value only late, in the storage merge, with an internal unmarshal error. A shared check now refuses a string, number, boolean, array or null up front on every create and update path with invalid --metadata: must be a JSON object; {} is still accepted. null is the one value update handled differently before: the storage merge accepted it as a no-op and exited 0, so bd update --metadata null now exits 1 where it used to silently change nothing. Values inside the object stay opaque, so a string value that looks like JSON is kept as written.

  • bd mol ready --gated no longer refuses under --proxied-server when BEADS_MAX_ROWS is set (#6293). The proxied capability front door keyed its rules on cobra's leaf name, and bd ready and bd mol ready --gated share the leaf "ready", so the row cap refusal written for the former also refused the latter — a proxy-supported command with no --max-rows flag to refuse. Rules are keyed on the full command path now.

  • bd init no longer tells you not to run bd init when it refuses a foreign database (#5558). When bd init opens a server database that already exists and belongs to a different project — for example when --server-host/--server-port/ --database point it somewhere other than this workspace's own server — it correctly refuses with PROJECT IDENTITY MISMATCH, but the refusal was the one written for ordinary opens, ending in "Do NOT run 'bd init'". The init-time refusal now names the database and points at remedies that work from there: point bd init at this project's server (--server-host/--server-port) or an unused database name (--database), or run bd doctor --fix / bd bootstrap if metadata.json is the stale side. Only bd init's own open gets this wording; every other open, including the library API, bd doctor --fix and bd bootstrap, keeps the existing message.

Changed

  • bd init refuses to initialize directly in your home directory (#4635). bd init sets up the directory it runs in as a project: it runs git init there unless it finds a repository, and installs agent files — CLAUDE.md, AGENTS.md, .gitignore, .claude/, .codex/, .cursor/, .agents/. In your home directory most of those are your agents' user-wide settings, so bd's hooks ended up running in every project, and init did this silently under --non-interactive, which engages automatically whenever stdin is not a TTY (so: whenever an agent runs it). A project is virtually never the home directory itself, so this now refuses with exit code 15 (ExitHomeDirRefused; 13 and 14 already mean a lost conditional-write race and a migration freeze), documented at docs/recovery/init-safety.md#init-home-refused.

    Deliberately narrow: it fires only when the working directory is exactly your home directory, your home is not itself a git repository (tracked dotfiles are unaffected, but a home that only sits inside an enclosing repository, or a GIT_DIR inherited from the environment, does not count), and you did not name a BEADS_DIR yourself. A subdirectory of home is an ordinary project and is never refused. If you do name a BEADS_DIR from an untracked home, init sets up only that workspace: no git init, no agent files or .gitignore in your home directory, and nothing in a repository your home sits inside (no hooks, beads.role, Dolt remote taken from its origin, or bootstrap commit) unless a flag such as --stealth or --setup-exclude asks for it. --force and --reinit-local do not override the refusal — they bypass the local data-safety guard, which is a different question from "is this the right directory". The guard runs first in init's RunE, ahead of every path with an effect, including the proxied dispatch and the store-opening --reinit-local pre-checks.

  • The managed shared Dolt server now opens its configured remotesapi listener (#6020). dolt.remotesapi-port (machine-global user config) and BEADS_DOLT_REMOTESAPI_PORT now reach the launched dolt sql-server (as --remotesapi-port in flag mode, remotesapi.port in generated-YAML mode), and doltserver.Restart applies a listener change under one lifecycle lock. Zero (the shared-mode default) still means no listener.

    This changes the meaning of a pre-existing knob: BEADS_DOLT_REMOTESAPI_PORT previously only informed federation doctor checks; a managed server launched with it set now actually opens that listener. A server still running from before the setting appeared refuses bd dolt start with a bd dolt stop && bd dolt start remedy, while auto-start keeps serving SQL and prints a warning until the server is restarted.

    Security caveat: Dolt binds the remotesapi listener on all interfaces and serves it unauthenticated, unlike the managed 127.0.0.1 SQL listener. Only enable it where a firewall, private interface, or tunnel bounds who can reach the port.

  • bd mol pour and bd mol wisp now reject a --var name the proto cannot consume, instead of dropping it silently. An unusable name was accepted and ignored, which turned a typo in an optional var into a conditional step that quietly never appeared, or a defaulted var that quietly kept its default - with a successful exit and no output to suggest otherwise. A name is accepted if any proto being poured declares it, references it as a {{handlebar}} in any field the pour substitutes - the prose fields, assignee, labels, a gate's await_id, and every string value in its metadata - references it from a step condition or from a step that condition removed, or uses it as a {name} placeholder that a standalone type = "expansion" formula substitutes as it builds its steps from [[template]]. The last three matter because the condition filter and the [[template]] expansion both run before the cook, and nothing in the cooked subgraph records the names they consume or drop; collecting them ahead of the filter also keeps a name's validity a property of the formula rather than of another var's value. Anything left over cannot affect the pour, so it is reported: unknown variables: has_spke (available: has_spike, story). A var belonging to an --attach proto still passes, as do a proto's documentation handlebars. The refusal applies only where the declared-var set is fully known, which means a proto cooked from a formula: a proto loaded from the database (bd cook --persist output, and every --attach proto) no longer has its [vars] declarations, so a declared-but-unreferenced name is indistinguishable from a typo and nothing is refused for that pour rather than risk breaking one that works. Out of scope for now: bd mol bond --var and bd mol seed --var go through formula.ValidateProvidedVars, which only iterates declared vars, so they still drop an unknown name silently. Also fixed alongside: the missing-var hint printed --var =<value> when there was no missing var to name.

  • A formula with a waits_for gate and no spawner to wait for is now rejected, and an invalid formula is no longer reported as not found. A gate step infers its spawner from needs[0], so with nothing to infer from, cooking stamped the gate:<value> label and emitted no dependency edge at all - a gate bead that never blocks and never resolves. bd create already refused --waits-for-gate without --waits-for; the formula path was the outlier. Formula.Validate now enforces the same rule. Separately, a validation failure is distinguishable from "this name is not a formula": callers that fall back to a proto/issue lookup previously swallowed the real error and reported not found as issue or formula, so an invalid formula looked like a typo. bd mol bond --dry-run also resolves through parser.Resolve now, so it resolves and validates the same way the real bond does rather than previewing a bond that then errors on those checks. Later cook transforms (inline expansions, expansions, aspects and condition filtering) are still only exercised by the real bond.

  • bd prime describes the memory split instead of prohibiting MEMORY.md (#6111, refs #5169). The guidance line used to read Do NOT use MEMORY.md files — they fragment across accounts. That warning predates harnesses that ship a first-party, machine-local memory whose index file is also named MEMORY.md, so in a workspace running both, prime told the agent every session not to use a file the harness was actively maintaining. The line now says what each store is for: durable project facts go in bd remember, because per-tool memory files fragment across accounts; per-operator preferences stay in the harness's own memory. Output text only, no behaviour change. The same sentence is updated in its sibling renderings — the MCP/minimal context, the generated AGENTS.md sections for the minimal and Codex harness templates, and the README snippet — so bd init and copy-paste do not re-introduce the prohibition.

  • BREAKING: unforced bd update --notes over existing notes is now refused, not warned (#5946). Replacing existing non-empty notes with different non-empty content — the blind clobber that destroys another agent's handoff notes — previously printed a warning after succeeding; it now writes nothing and exits 1. The same fence turns PATCH /v0/beads/issues/{id} and batch:apply from 200 into 409 notes_overwrite_refused when patch.notes would do the same, on both backends and both batch legs. Migration: pass --force (CLI) or force_notes_overwrite (API) to overwrite deliberately — the CLI then prints the old warning as an audit trail — or preserve history with --append-notes / bd note.

    Two edges are deliberate. An explicit clear is never fenced behind --force: --notes "" (or an API patch to the empty string) does not trip this refusal, because every sibling text field clears the same way, the pattern the fence exists to stop is an agent writing its own content over someone else's, and the refusal's advice — force, or append — is meaningless for a clear. (The CLI gives the clear its own verb, --clear-notes — see the next entry — but never gates it behind --force.) And --force is no longer mutually exclusive with --if-assignee, since a guarded notes overwrite has to be able to say both. This widens what --force reaches under the guard: its assignee half stays suppressed (a transfer under --if-assignee authorizes only through the matching CAS, never the force bypass), but its close-policy half — closing despite open children or a live blocker — now applies there too, a combination the CLI previously rejected outright. bd edit is unaffected: it pre-fills the editor with the current notes, a sighted edit rather than a blind clobber.

  • bd mol bond --ref no longer deadlocks the molecule it bonds into, and no longer accepts --type conditional. A --ref arm is nested inside its target and its hierarchical ID records that, so an ordering edge onto the target was unsatisfiable in both directions: the arm waited for the molecule to close, and the molecule could not close while it held an open arm. bd ready went empty with nothing to point at. For a sequential bond (the default) that edge is now dropped, so the arm carries no ordering and is ready as soon as it is spawned - containment is a nested arm's only satisfiable relationship to its container. A conditional edge has no such degradation: it means "run only if the target fails", so dropping it would run the arm unconditionally, turning a deadlock into a silent false dispatch. --ref with --type conditional is therefore refused outright (including under --dry-run and on the proxied route); omit --ref to bond a conditional arm as a sibling, which is satisfiable and keeps its edge.

  • bd mol ready --gated prints a dispatch command that exists. The hint named bd sling, which has never been a command in this binary, so the one actionable line in the output could not be run. It now prints bd assign <ready-step-id> <agent>, and says which listed molecule the example belongs to when more than one is waiting.

  • bd purge keeps closed beads a live bead depends on. A closed bead that a not-done bead depends on through parent-child, tracks or blocks is no longer purged (reported as live_dependent_skipped), so a purge that used to delete a closed molecule root under a live step now leaves it. See the --wisps-plane entry under Added.

  • bd status now discloses the counted rows a default bd list will not show. Total Issues counts the database; bd list shows a filtered view of it, and the two disagreed silently — an operator reconciling a total against a listing found rows they could not account for. A new line names the breakdown and the flags that reveal each part: Not shown by bd list: 2 gates (--include-gates --all), 1 template (--include-templates --all), omitted entirely when there is nothing to disclose. The counts describe the database, every status included, which is why the remedy names --all beside the type flag: the default listing hides closed rows independently of what a row is, so a closed gate is counted here and the type flag alone would not show it. gate_issues and template_issues join the --json payload and the OpenAPI Statistics schema, additively; they are broken out of TotalIssues rather than removed from it, so the buckets still do not sum to the total in either direction. bd status --assigned gets the same breakdown, which it needs for the same reason against bd list --assignee. Not yet covered: durable rows carrying a configured infra type (GH#6439), which need the workspace's types.infra set that the portable count has no seam to read.

  • Proxied-server refusals now say why they refuse. The JSON a refused command prints gains a reason field next to the existing code, error and mutates: design for a refusal that is expected to stay (shared history, multi-repo routing, destructive admin, strict --readonly) and unimplemented for a capability gap with a named owner. Adding reason left every existing code, message and exit status as it was, and reason is absent on refusals that report a runtime state rather than a policy, so existing consumers are unaffected. The policy behind it moved into one registry (cmd/bd/capability_registry.go) that every command must appear in.

  • A command with no proxied-server route now fails with a typed error (proxy.store.unrouted) instead of the bare string proxy server store should be uow provider.

  • bd backup restore --json prints a result object on every topology ({"restored": true, "source": "<dir>"}). It previously printed nothing at all, which left a caller unable to tell a completed restore from a silently skipped one. This applies to direct/embedded workspaces too, not only proxied ones.

  • bd backup init, sync, remove and restore no longer print a usage block after a failure. They now silence cobra's own error and usage rendering like the rest of the CLI, so a failure is one error line on stderr instead of the message followed by Error: exit code 1 and the full usage text. Exit statuses are unchanged; this applies on every topology.

  • bd label remove and bd label add no longer claim a no-op edit (#5988). Removing a label the issue never had printed ✓ Removed label 'x' from <id>, and adding one it already had printed ✓ Added label, so a real edit could not be told from a no-op. Such a label is now reported as • Label 'x' was not on <id> / • <id> already has label 'x', with JSON status unchanged instead of removed/added; a multi-label edit reports each label by what actually happened to it. The exit code is still 0, so idempotent callers are unaffected.

  • Proxied --repo and row-cap refusals are typed again (#6293). bd create --repo under --proxied-server answers --json with the stable {"code": "proxy.repo.unsupported", …, "mutates": false} on stdout rather than prose on stderr, and bd ready --claim --max-rows N refuses with the same proxy.max_rows.unsupported shape as bd ready --max-rows N: --claim does not exempt a row cap the proxied route cannot enforce either way. bd create is the only command that registers --repo, so it is the only command this refusal applies to.

  • bd ready --gated runs under --proxied-server with a row cap set (#6697). It was refused whenever --max-rows or BEADS_MAX_ROWS resolved to a positive cap, while bd mol ready --gated — the documented alias for the same gate-resume scan — was not, so one command line behaved differently under its two spellings. The gated arm lists molecules whose gate closed rather than ready rows and threads no cap on either route, so it no longer consults one; a cap passed alongside --gated is ignored, matching what the direct route has always done. Every other bd ready arm still refuses an active cap, --claim included.

  • The proxied bd admin compact refusal names the command it means (#6293). The message is now "only 'bd admin compact --dolt' is supported in proxied-server mode"; it read "only 'compact --dolt'", which names the root bd compact — a different command, with no --dolt flag and its own proxied route. Scripts matching the old text need updating.

  • bd query no longer silently stops at 50 rows when its output is piped (#6229). bd list has treated piped stdout as unlimited since GH#4094, but bd query read its --limit default of 50 straight through, so bd query '...' | consumer dropped every row past 50 with no hint. An unflagged bd query now resolves through the same policy as bd list: an explicit --limit wins, piped stdout is unlimited, agent mode on a terminal gets 20, and a terminal gets 50.

  • bd preflight honors the json config default (#6293). json: true in the config file selects JSON output for bd preflight, as it already did elsewhere; previously only the explicit flag was honored.

  • BREAKING: bd update --notes "" is now refused; clear with --clear-notes (#6021). An empty --notes previously wiped the whole notes field at exit 0 behind the same ✓ Updated receipt as a successful write — and the empty string is exactly what a dead command substitution (--notes "$(cat missing.txt)") collapses to, the accident behind every recorded loss in the report. At the flag layer the accident and the deliberate clear are the same bytes, so the empty value is refused unconditionally (exit 1, naming the alternative) and the deliberate clear gets its own verb: bd update <id> --clear-notes, mutually exclusive with --notes and --append-notes. A verb rather than an --allow-empty-notes opt-in because what needs authorizing is an intent, not an input path — the description guard's --allow-empty-description unblocks stdin/file plumbing, which notes does not have — and because an opt-in passed habitually would silently disarm the guard, while a habitual --clear-notes fails loudly on the next real update. --force does not bypass the refusal (the overwrite fence answers a different question), --append-notes "" remains a safe no-op, and the HTTP API is unchanged: patch.notes: "" still clears — a JSON payload is a deliberate construction, not a shell substitution.

  • --notes help no longer reads as additive (#6272). bd update --help described --notes as "Additional notes" and the bd prime text said "Add supplementary notes" — both read as appending, so agents following the injected guidance faithfully clobbered prior context. The update help now names the operation ("Replace the notes field …") with the --append-notes contrast at the point of use, and the base/create help is field-named like --description and --design, implying no merge behavior.

  • bd gate check resolves bead gates whose target lives in a prefix-routed rig (#5859). After a local miss, the evaluator follows the target bead ID through routes.jsonl and reads the owning store without writing to it. This covers explicit gate checks in embedded, server, and proxied-server command paths; the legacy <rig>:<bead-id> await value remains accepted for compatibility.

Fixed

  • bd gate check no longer reports an unreadable store as "pending". The bead arm of the check dropped the lookup error on the floor, so with dolt down (or any backend or transport failure on the awaited bead's read) every bead gate printed as still waiting and the command exited 0 — the same output as a healthy, genuinely pending gate. A read of the workspace's own store that fails for any reason other than not-found is now an error row (✗ <gate>: error checking - ...), counted in the summary, and bd close on such a gate keeps refusing (could not check bead gate, --force to override) rather than letting a dead store read as satisfied. On the classic and proxied routes alike, bd gate check now exits non-zero whenever any gate it checks (gh, timer, or bead) could not be checked or closed. A bead in a prefix-routed rig whose store cannot be read stays pending, with the routing failure in its reason (cannot confirm the awaited bead is gone); for a missing bead, see #6395.

  • bd gate check resolves a bead gate whose awaited bead was deleted (#6395). A bead gate waited for its target to close, so deleting the target left the gate pending forever and its step blocked. The first bd gate check that finds the target in this rig's own store now records the sighting in the gate's metadata (await_seen), and a later check that no longer finds it there resolves the gate (awaited bead <id> no longer exists). An await ID that no check has seen, such as a typo or a bead deleted before the first check, stays pending with a diagnostic instead of unblocking the step: correct the await ID or close the gate with bd gate resolve. A miss outside this rig's store (a target whose prefix routes to another rig, or a contributor auto-routed store) does not prove the bead gone, so the gate stays pending and the reason names the cause. bd close on a bead gate follows the same rule: it no longer needs --force once an earlier check saw the awaited bead and the bead is gone from the gate's rig, with no route pointing elsewhere. bd rename and bd rename-prefix point a bead gate at its bead's new ID, sighting included, so a renamed bead does not read as deleted, even to a check or bd close that read the gate before the rename; a rename that fails partway leaves the gate pending without a sighting, never resolved. A rename by an older bd does not move the gate: if a check had already seen the bead, a later check or bd close treats that rename as a deletion and closes the gate while the bead is still open, so upgrade every client that renames beads together with the ones that check gates.

  • An epic that blocks on its own children no longer hides them from bd ready (#6506). A parent-child edge now propagates only a parent's exogenous blockedness — blockedness whose cause lies outside the parent's own subtree — so the close-gate idiom, a parent carrying blocks edges onto its own children (or grandchildren) so it cannot close before them, stops darkening the very work it is waiting for. The parent itself stays blocked, and a close gate does not unlock the work under an epic that is itself blocked from outside: a sub-epic blocked BOTH by its own children AND by an exogenously blocked ancestor still darkens its children (#6601). "Inside my own subtree" is decided to a fixed depth of four parent-child levels — epic → sub-epic → leg → task — so a parent that blocks on something deeper than that still darkens its whole subtree, exactly as it did before this fix; that depth is a measured trade (the bounded walk is a fixed chain of index probes, where a recursive one cost 7.5x per evaluation) and raising it is a one-constant change. Existing stores repair on the next write that touches the hierarchy, or with bd doctor --fix.

  • bd dep add no longer explains its refusal of a blocking edge onto your own descendant by claiming the block would cascade down and never clear — it does not, as of the fix above. The refusal stands, and now names the sanctioned way to say it: a waits-for gate over the children.

  • routes.jsonl prefixes containing a hyphen now route (#5048). Prefix routing cut the bead ID at its first - and required an exact match, so a route such as claude-os- could never match claude-os-76l and the lookup fell through to "not found". Routing now picks the longest route prefix the ID starts with. Single-hyphen routes behave as before; where both hq- and hq-cv- are configured, hq-cv-* IDs now take the hq-cv- route instead of hq-.

  • A dotted config key now round-trips: what bd config set writes, bd config get and bd's own readers find (#6574, #6594, bd-zj95). Setting a key like sync.remote or dolt.host into a config.yaml with no matching section appended a key whose name contained the dot — a top-level sync.remote: "..." — instead of nesting it under sync:. Viper finds a key spelled that way; the direct reader GetStringFromDir splits on the dot and walks nested mappings, so it does not. The value was set and invisible at the same time depending on which reader asked, and bd init --remote in a fresh workspace was the ordinary way in: the remote was recorded and then not found. New dotted keys are written nested now. An existing flat spelling is updated in place for compatibility with other writers of the shared config file — rewriting that one line and nothing else, the comment on it included — and bd's direct reader recognizes both forms with the same precedence as Viper. bd doctor warns when both forms coexist. bd config unset removes the nested form — it only ever matched the flat spelling, so unsetting sync.remote silently left the remote live. Keys the write does not own, including dotted ones somebody else put there, are left exactly as they are.

    Four things that used to succeed now report an error instead, because each one produced a value no reader could see, or claimed to remove a value it had left live. On bd config set: a key under a parent that already holds a value (sync: enabled, then bd config set sync.remote), and a key in a file whose top level is not a mapping. On bd config unset: a key inside a flow-style mapping (sync: {remote: "..."}), which bd matches by line and so cannot edit, and a key whose value is a block scalar (remote: |, in either spelling), where commenting the key out would leave the body behind as a value of its own. Unsetting a key that is not set remains a successful no-op in every shape.

  • A self-hosted GitLab no longer re-creates its own issues on every push (#6735). The tracker decided whether a stored external_ref was one of its own by looking for the substring "gitlab" in the URL, so an instance whose hostname does not contain it — https://nova.teachx.ai, say — had its own issue and work-item URLs classified as foreign. bd gitlab push then took the create-not-update path for issues that were already synced, and pull reconciliation and conflict detection skipped them. Such a URL is now also recognized when its host, and base path for a GitLab served from a sub-path, match the configured gitlab.url.

    Recognition only widens: every ref the substring check already claimed is classified exactly as before, so no existing link changes meaning. Two limits are worth knowing. Matching is host-scoped, not project-scoped — with GitLab at the host root, a ref pointing at another project on the same host now counts as this tracker's, as has always been the case for gitlab.com. And the host is compared literally, so gitlab.url must name it the way GitLab spells it in the web_urls it returns; writing an explicit default port (https://host:443) matches none of them and leaves the old behavior in place.

Changed

  • bd ready --json answers a capped page and its total in one pass, and storage.Storage gains a required GetReadyWorkWithCountsAndTotal method (#6731). A full capped page used to trigger a second counting pass in its own transaction, re-running the expired-defer sweep and every wisp/deferred probe; the total now comes from COUNT(*) OVER () on the page's own query inside the same read transaction. Two behaviors improve with it: the --json total honors external-dependency exclusions (the old second pass counted externally blocked issues, so text and --json could report different numbers), and the page and its total now come from one snapshot instead of two, so "Showing N of M" can no longer be internally inconsistent under concurrent writes. One contract change: a total that fails now fails the command, where it previously could not — the total is taken from the page's own query, so the page would have failed anyway.

    Out-of-tree storage backends must implement the new method to compile. Every in-tree implementer is updated, but internal/storage/backends is an extension seam whose registrants are supplied by a downstream distribution and return storage.DoltStorage directly. A registrant must implement GetReadyWorkWithCountsAndTotal(ctx, types.WorkFilter) ([]*types.IssueWithCounts, int, error): the page identical to GetReadyWorkWithCounts, and the total identical to the number CountReadyWork answers for the same filter (which ignores Limit), resolved in the page's own read transaction. There is deliberately no optional-interface fallback — a store that cannot size the ready set should fail to compile rather than silently fall back to an unbounded query.

  • storage.Storage gains a required SearchIssueSummaries method (#3458). It is a narrow-projection variant of SearchIssues for list-shaped rendering: it returns []*types.IssueSummary (exported as backend.IssueSummary), which carries the list columns, labels, and the four wisp-plane markers but none of the TEXT/JSON columns. Nothing in bd calls it yet, so no command's output changes. Consumers that only call the interface are unaffected; any external type that implements it (a custom store, mock, or proxy, or an internal/storage/backends registrant) must add the method to compile: SearchIssueSummaries(ctx context.Context, query string, filter types.IssueFilter) ([]*types.IssueSummary, error). It honors SortBy/SortDesc and SkipLabels exactly as SearchIssues does, IncludeDependencies is a silent no-op (a summary has nowhere to put dependency records), and wisps are merged in unless SkipWisps is set, with their markers set so a wisp row stays distinguishable from a durable issue.

  • Push --dry-run now honors --create-only (#6337). The sequential tracker push previewed "Would update" for every already-linked issue, even though a real --create-only run skips them, so the preview over-reported updates and under-reported skips. This affected bd jira sync and bd linear sync with --push --create-only --dry-run. The dry-run now applies the same --create-only gate as the real run. Linear's dry-run also no longer counts an issue its batch filter skipped (create-only, a ShouldPush hook, parent, type, or conflict) as skipped twice. Notion's batch dry-run was already correct.

  • bd config unset no longer reports success while the key stays set, and now names where the unset actually landed. A key present in config.yaml but not claimed by IsYamlOnlyKey was deleted from the database only, so it stayed effective; it is now cleared from both, and the reported location - Unset <key> (in database, config.yaml) - is derived from what the writes actually changed rather than from a pre-check of viper's merged value, which counted defaults and environment variables as present in the file. A yaml-only key that was not in the file prints <key> was not set in config.yaml instead of claiming a write. A workspace with no project config.yaml at all is an answer rather than a failure, so a database-backed unset there no longer deletes the row and then exits non-zero. Unsetting a key whose value is the indented block beneath it (a mapping or a list), which used to comment out the key line and orphan the block into a config.yaml that no longer parsed, is now refused alongside the flow-style and block-scalar shapes above. An unset also keeps the file's trailing newline, and an unset that changes nothing now leaves the file byte-identical - on a CRLF config.yaml it previously reported a write for a key that was never there and rewrote every line ending to LF. The stderr side-effect hint is gated on the write too, so a no-op no longer announces a removal it did not make, and the --json payload for a yaml-only key carries a changed boolean so the machine branch can tell an absent key from an unpopulated field.

[1.3.0] - 2026-09-15

The first tested release off main since the 1.1 line. [1.2.2] was a recovery release that re-shipped the v1.1.2 code under a higher version number, so a v1.2.2 user is running 1.1-era code and meets everything below at once: the changes listed here, plus the [1.2.1] changes that release deliberately withheld. Read the upgrade notes before installing.

Upgrade notes

On an embedded or local store, the first invocation migrates your schema, in place, from v53 to v66. A v1.2.2 (or any 1.1.x) database sits at main-series schema v53; this binary knows v66, so the first command that opens the store applies 13 main-series migrations, 0054_add_lease_columns through 0066_add_events_journal_actor. Two of those passes rewrite rows rather than just reshaping tables — the aux-row id rekey and the events dolt_ignore flip described under Changed — so on a large store the first invocation is noticeably slower than the ones after it. It is crash-resumable and picks up where it left off, but do not interrupt it if you can avoid it.

A shared Dolt sql-server is never auto-migrated (#5920, #6048): migrating it promotes the schema for every client at once, so it waits for explicit consent via bd migrate schema. Remote-backed stores keep their existing designated-migrator gate. Upgrade all of that server's clients first — see Shared servers.

The counter restarts partway through, and that is not a loop. The clone-local (dolt_ignored) series runs after the main one, through the same printer and its own numbering, and it moves 0011 → 0026 on this upgrade. So the run is about 28 migrations, not 13, and what you see on stderr is 13 lines counting up to 0066 followed by 15 lines starting again at 0012:

Applying migration 0065: widen_wisp_comments_text…
Applying migration 0066: add_events_journal_actor…
Applying migration 0012: create_leases…      ← clone-local series, not a restart

A counter that jumps backwards is the signature operators kill runs over. Let it finish.

Progress prints only when stderr is a terminal. Piped and CI runs see nothing at all, which is deliberate — a silent-looking CI upgrade is not a stuck one.

Back up first, with the binary you have now. Take the backup before you install 1.3.0. Under the new binary bd export triggers the auto-migration before it exports, so a snapshot taken afterwards is a post-migration snapshot and cannot protect you against the migration going wrong. On a remote-backed store, finish syncing with the old binary too: once 1.3.0 is installed the pending-migration gate refuses bd dolt push and bd dolt pull as well, not just bd migrate.

# with your CURRENT bd, before installing 1.3.0:
bd dolt push                                                   # remote-backed stores only
bd export --all -o .beads/backup/pre-1.3.0-$(date +%Y%m%d).jsonl

A JSONL export is cheap, issue-complete, and importable by any bd version. If you want a Dolt-native snapshot that keeps history and config, configure a destination and sync it — bare bd backup takes no backup, it is a command group that prints help and exits 0:

bd backup init <path-or-dolthub-url>   # once, to configure a destination
bd backup sync                         # take the snapshot

Upgrade every client that shares a store, together. The forward schema-skew guard means an older co-resident binary — a second bd earlier in PATH, a long-running bd serve, another clone's cron job — refuses a database migrated past what it knows, rather than proceeding blind. That is the guard working, not a bug, but it makes a mixed-version fleet a broken fleet: one machine running 1.3.0 takes the whole store forward and every 1.2.2 client stops. Run which -a bd after installing, and on a remote-backed store follow the existing designated-migrator procedure (one clone migrates and pushes; the rest pull or re-clone). See Upgrading for the per-install-method recipes and the multi-clone flow.

If you need to go back, the rollback is a schema-cursor rollback, not a downgrade of the data: the procedure is written up in the recovery runbook (repo copy: docs/recovery/accidental-1-2-1-release.md). Its worked example is the v53↔v65 case from the accidental 1.2.1 release; the steps are the same for v66, with the version numbers adjusted.

Breaking changes for v1.2.2 users

This is a highlights list, not the complete set. A v1.2.2 user is crossing two releases at once, and the breaks below are the ones most likely to stop a script or a service. The full set is the [1.2.1] section further down plus the Changed section here — read [1.2.1] in full before upgrading, since [1.2.2] withheld it and v1.2.1 itself was pulled, so nobody on the supported line has seen it.

After upgrading, bd upgrade review prints exactly the entries between the version you were running and this one. (Prefer it to bd info --whats-new, which dumps the entire release history.)

Carried from [1.2.1] — never shipped to v1.2.2 users:

  • bd update --status <done-status> now enforces close policy. Moving an issue to closed (or any configured done-category status) via bd update, bd batch update, or the issueops facade refuses when the issue has open children or a live direct blocker, matching bd close. Override with bd update --force (update <id> status=closed force=true in bd batch; UpdateRequest.ForceClosePolicy for facade consumers). An unforced refusal rolls back the entire batch.
  • bd search includes closed issues by default (bd-t5yex). The dominant query is "was this already found/filed/fixed?", where excluding closed issues produced a false "no". Narrow with --status open to get the old behavior. bd list keeps its open-only default.
  • The no-ID "last touched issue" fallback on bd update / bd close is interactive-only (bd-m00pb, #4839). A scripted bd update $ID … with an empty $ID now refuses in argument validation instead of mutating whatever was touched last. The fallback requires a terminal stdin; BD_NON_INTERACTIVE=1 and CI=1/true also disable it. Set BD_LAST_TOUCHED_FALLBACK=1 if a script genuinely relied on it.
  • bd human list hides done/frozen and pinned beads by default, and validates --status (#5332). It previously passed --status through unvalidated and showed closed beads. A --status typo is now an error rather than an empty list. Every bead type still shows.
  • bd dolt push and bd sync no longer adopt a git-origin-derived Dolt remote without consent (#5068). Both used to silently derive a remote from git remote get-url origin, persist sync.remote, commit that config change under your git identity, and upload the full issue history — which published a whole issue database to a public origin on a command the user believed targeted an already-configured remote. Adoption now prompts (defaulting to no) and fails closed non-interactively. --yes/-y consents ahead of time.
  • bd --readonly serve is refused instead of binding a server that cannot do what it advertises. On a Dolt SQL-server workspace the flag was previously a silent no-op and the server came up fully writable, so anything scripted as a "safe" read-only server does not start after this upgrade — drop the flag. Worth checking before you restart a long-running bd serve as part of the fleet upgrade above.
  • bd config list and GET /v0/beads/config no longer enumerate the kv. plane, which is where bd remember memories live. That closed an unauthenticated GET /v0/beads/config handing out every stored memory. A script that read memories out of bd config list now gets empty output rather than an error; use bd kv / bd remember instead.
  • The published backend package drops orphan handling (bd-gwryr). backend.OrphanHandling and its four constants, the storage originals they aliased, BatchCreateOptions.OrphanHandling and issueops.CheckOrphan are removed. They never did anything except cost a query, but a Go consumer that names them no longer compiles.

New in 1.3.0 (full entries under Changed):

  • An explicitly configured Dolt server port now outranks the ambient BEADS_DOLT_PORT environment variable.
  • Actor matching decodes an exact -- run to / instead of collapsing it to a generic separator, so gastown--mayor matches gastown/mayor and stops matching gastown__mayor.

Added

  • The events journal records WHO performed each mutation. bd_events_journal gains an actor column (migration 0066 plus its ignored-series twin 0025, so upgraded workspaces and fresh clones converge on the same shape), stamped inside the mutating transaction with the same identity the audit-events table resolves; on a comment row it is the comment's author. bd events tail / bd events export JSON gains an additive, omit-when-empty actor field — empty means the path had no actor (derived maintenance, deletes, rows older than the column), never a user. The input for same-field / different-actor conflict measurement over the journal.

  • bd warns when it stores a label containing a space (#5813). Such a label is a legitimate thing to ask for — -l 'good first issue' — and bd stores it as asked rather than re-splitting a boundary the shell already decided. But it is also exactly what a missed comma looks like, and that case cannot be told apart from the deliberate one at the point of writing. A line on stderr catches it at the keystroke instead of months later: ⚠ Stored "auth backend" as ONE label — it contains a space. It is advice, not an error; stdout is untouched, so --silent still prints just the ID, and --quiet silences it. Not emitted for --remove-label, since removing such a label is how the damage gets repaired.

  • bd doctor gains a Label Whitespace check for finding damage already in a database (#5813). It reports labels that contain whitespace or differ from their trimmed form, and bd doctor --fix repairs them. The check is warn-only: a database with legacy damage still exits 0. As with the rest of bare bd doctor, it does not yet run in embedded mode (GH#3794).

  • bd stale and bd blocked gain --label, --label-any and --exclude-label (#5822), with the same meanings and the same flag names they already have on bd list and bd ready: --label is AND (repeatable, must have all), --label-any is OR, --exclude-label drops anything carrying one. They were the two listing commands with no label-scoped output at all, so a theme-scoped stale sweep had to pull the whole list as JSON and post-filter it on the labels array. Filtering happens in SQL ahead of LIMIT, not on the hydrated rows, so --label x --limit 10 returns ten matching issues rather than whatever survives filtering the first ten. No --theme flag: beads has no theme concept, themes are labels here, and --label theme:x is the native way to say it.

  • New *.gate.lock files appear next to and inside .beads (#5046, #5093). A two-level cooperative gate now serializes the operations that cannot safely overlap — ordinary commands take it shared, maintenance like bd backup restore takes it exclusive. It is flock-based, so the file is the lock's name, not its state: the gate is created once (mode 0600) and deliberately never deleted, because deleting it is how two processes end up holding two different locks for the same resource. Each gate sits beside what it guards, never inside it: the workspace gate for <dir>/.beads is <dir>/.beads.gate.lock in the project root, and the physical-root gate is .beads/embeddeddolt.gate.lock in the default embedded mode (.beads/dolt.gate.lock under a server layout; under the shared server it lives beside ~/.beads/shared-server/dolt and never enters the workspace at all; a remote Dolt host has no physical gate). Both are covered by the *.gate.lock* pattern bd doctor maintains in the project and .beads gitignores, so an upgraded workspace picks up the ignore rule on the next bd doctor --fix; a workspace with no .beads directory is not gated at all, rather than scattering lock files into whatever directory bd was run from.

  • sort on GET /v0/beads/issues. The listing served one order — (created_at DESC, id ASC) — because the cursor is a keyset position in it, and the spec said so in its own words: "the sort order is welded to the cursor contract, and a new order needs new surface." The cost of that welding fell on every client that wanted bd list's ordering, because the only way to get it was to page the whole result set and re-sort locally. Measured over a 1400-row store at the 200-row page an HTTP client actually uses, that is 7 requests plus a client-side comparator; ?sort=priority is 1 (TestProxiedServerListSortRetiresTheWalk, which runs both strategies against a real bd serve and checks both against bd list --json row for row). It is the one cost on this surface that got worse as a project grew.

    Two values, and the set is closed. created is the existing order, now spellable. priority is (priority ASC, created_at DESC, id ASC) — bd list's flagless ordering, which is also what bd list --sort priority produces, so one served order retires the walk for both. The other seven orders bd list --sort takes are not offered: each value here is a cursor contract needing a key proven total, and id is a natural-numeric order no database expresses, updated moves on every write, closed is nullable, and status/title/type/assignee are mutable and unindexed.

    Absent sort still means created, permanently. It is the compatibility contract for every client written before the parameter existed; changing it would alter which rows a truncated page contains, with no error to notice it by. A server that predates the parameter answers param: "sort" with reason: "unknown_parameter", which is the per-parameter capability probe a client dispatches on to fall back.

    The cursor now carries its order, and a mismatch is refused. This is the half that makes the parameter safe rather than merely useful. The token was base64 of {t,i} with no binding, so a created-order cursor replayed under sort=priority would have decoded perfectly and been read as a position in a different total order — a page that both skips and duplicates rows, served with a 200, undetectable by the client. Tokens are now v2 with an explicit order member and decode refuses any token whose order differs from the request's, as invalid_cursor (documented recovery: restart paging). Outstanding v1 tokens stay readable as the created-order positions they are, so no traversal in flight has to restart. This does not contradict "the token carries no filters": filters select the set, the order decides what the position means.

    priority is a mutable key, which created_at is not, and the spec says so: under created only new rows move relative to a walk, while a priority update moves an existing row too, so it can be seen twice or missed. That is the already-documented "a cursor pins a position, not a snapshot" caveat reached by a second route, not a new class of error — unchanged data never skips or repeats under either order.

Changed

  • Shared Dolt databases (sql-server mode) no longer auto-apply schema migrations on a version bump. Migrating a shared database promotes the schema for every connected bd client at once and locks out clients still on an older binary, so bd now refuses without explicit consent: run bd migrate schema once after upgrading every client, or set BD_ALLOW_REMOTE_MIGRATE=1 in scripted use. Reads keep working on the current schema meanwhile. Embedded (single-user) databases still auto-migrate silently, unchanged. (#5920)

  • Commands that cannot work under --proxied-server now refuse before any provider opens (#6293). bd doctor, bd backup, bd restore, bd diff, bd migrate, bd branch, bd conflicts, bd vc, bd federation, bd repo, bd flatten, bd sync, bd dolt push|pull|commit, bd dolt remote add|list|reset-data, and bd admin compact without --dolt return a typed refusal carrying a stable code (proxy.doctor.unsupported and friends) instead of failing partway through, or silently doing nothing. Two changes worth planning for:

    • bd doctor under --proxied-server used to print a note to stderr and exit 0; it now exits 1 with proxy.doctor.unsupported. A script that treated bare bd doctor as a no-op in this mode needs to stop calling it, or tolerate the non-zero status.
    • Under --json, a refusal is strict JSON on stdout ({"code": …, "error": …, "mutates": false}) rather than prose on stderr, so a wrapper parsing stdout gets a machine-readable answer and "mutates": false states that the refusal touched nothing.
  • The smart migration gate's "auto-migrate as safe first-mover" and auto-fast-forward arms are now embedded-only. On a shared server the gate always stops for consent, because neither argument can observe the other clients attached to that server. (#5920, #4516)

  • The newly exported Go packages (backend, backend/conformance, beadserrors, issueops, journalops, memoryops, schema, test/conformance) are marked EXPERIMENTAL: they are not yet covered by the project's compatibility promise and may change in a minor release. Pin an exact beads version and re-run the conformance suite on every bump.

  • The optimistic-concurrency token is now a JSON string — responses carry revision as an opaque decimal string and expected_version must be sent as that string. JSON-number tokens exceeded JavaScript's 2^53 integer precision. This member never shipped in a tagged release.

  • CI: migration-test and upgrade-smoke now seed and assert wisp-plane data (wisps, wisp deps/comments, leases, events, ignored-track cursor) across v1.0.1/v1.1.x/v1.2.2 → candidate upgrades.

  • --profile is now --cpu-profile, with no alias (#5126, bd-ugz). The persistent flag that writes a CPU profile is spelled --cpu-profile; the old --profile spelling is gone rather than deprecated, so it fails as an unknown flag instead of silently doing nothing. --profile is the conventional name for a named configuration, not for pprof output, and holding it hostage to a debugging aid nobody passes twice a year was the wrong trade. Scripts and aliases that profile bd need the new spelling.

  • Migration 0061 rekeys every events, comments, snapshot and compaction-snapshot row to a content-derived id (#5150, bd-ri8bd). Those four aux tables minted random UUIDv7 ids, which is fine for a single writer and wrong for replication: unversioned newest-wins replication has no merge pass, so two clones recording the same fact produced two rows that never converged. Ids are now UUIDv5 over a SHA-256 digest of the row's own content (frozen column order, plus the table name and a duplicate-disambiguating ordinal), so the same fact derives the same id everywhere and convergence is a property of the key rather than of a reconciliation step. New rows derive their id at insert; existing rows are converted by a one-time bulk rekey on the first open after upgrade, which is the main reason that invocation is slower than the ones after it. The pass is crash-resumable through a local_metadata sentinel and records completion clone-locally, so it runs once per clone and is not repeated on the next command. wisp_events and wisp_comments are deliberately excluded — they are clone-local and never merge. Any external reference to an event, comment or snapshot id taken before the upgrade will not resolve afterwards.

  • The events audit table is now clone-local (#5162, bd-red8u). Migration 0062 moves events onto the dolt_ignored plane in a self-committing drop/commit/recreate that preserves every row. Events are single-writer audit rows with low cross-machine value, and minting a versioned Dolt commit for each one was the dominant source of commit churn on a busy store. The consequence is the point: events rows no longer replicate. They keep full SQL durability locally and bd history <id> --events still reads them, but a pull no longer brings another machine's events over, and a push no longer sends yours. Events already committed before the flip stay in the remote's history until the next squash window. Comments stay versioned.

  • The interactions.jsonl audit sidecar is opt-in (#4688, fixes #4687). audit.enabled now defaults to false and bd init no longer creates the file, so status/assignee/priority field-change logging writes nothing unless you ask for it. Set audit.enabled=true (or BD_AUDIT_ENABLED=1) to restore the old behavior. The database-backed replacement is bd history <id> --events, which needs no sidecar. Anything that read .beads/interactions.jsonl on the assumption it would exist needs the config key set.

  • bd import refuses a redirected stdin rather than quietly ignoring it (#5171, bd-axluy). Bare bd import imports the default JSONL, and bd import - imports stdin; a bd import < file.jsonl looked like the second and behaved like the first, importing a different file than the one on the command line with no way to notice. With stdin redirected, no -, and no named source, bd import now errors and names both fixes. /dev/null is a character device, so a scripted bare bd import under a subprocess still imports the default file as before.

  • Auto-backup defaults to off under a Dolt sql-server (wy-zrmqr). The automatic backup used to switch itself on whenever a git remote existed, regardless of storage mode — which on one shared sql-server meant ~31 clients independently deciding to back up the same database, pinning its CPU. The default is now off in server, proxied-server and shared-server modes (and in nocgo builds, which are always server-backed); embedded mode is unchanged and still enables it when a git remote is present. An explicitly configured backup.enabled (or BD_BACKUP_ENABLED) always wins over the default, and bd config get backup.enabled now prints the effective value with its source — default (auto: off in sql-server mode) — rather than a bare false that could not be told apart from a configured one.

  • bd hooks install --chain and --force are accepted no-ops (#5284, bd-5vdt8). Managed BEGIN/END BEADS INTEGRATION marker sections made both flags meaningless: an install now replaces only the content between its own markers, so content outside them is always preserved (what --force used to negate) and an existing hook always keeps running alongside the bd section (what --chain used to request). The flags remain registered so existing invocations do not break, and --json still echoes them, but neither changes what is written. A hook with no markers is handled by shape: a legacy bd hook is replaced, and a foreign one gets a one-time <hook>.backup sidecar before the section is injected. Symlinked and git-tracked hook paths are refused outright, before anything is written.

  • Write commands refuse to run while a migration freeze marker is present (dc-6jaq). When a file named MIGRATION-FREEZE exists in the workspace directory, the working directory, or any ancestor of either — or at the path named by BD_MIGRATION_FREEZE_FILE, which is authoritative when set — bd create/update/close/remember/import and every other write command (~120 call sites, bd q included), plus the destructive bd init --reinit-local and bd bootstrap, print ⛔ workspace is frozen for migration naming the marker's path and exit 14, a distinct code scripts can branch on, instead of writing to a store mid-migration. Operators freeze a workspace by creating the file (optionally operator<TAB>RFC3339 timestamp<TAB>reason on one line, echoed back in the refusal) and thaw it by removing the file. Read commands (list, show, ready, …) keep working during a freeze, and bd's own maintenance stands down with them — version tracking, auto-migration, JSONL auto-import, Dolt auto-commit, auto-backup, auto-export and auto-push are all skipped, so a frozen store is never rewritten just because someone ran a read. A running bd serve is not gated: stop it before freezing. If bd cannot tell whether a marker is present, it refuses rather than assuming the workspace is open.

  • bd dep add names the implicit type=blocks default, but only to an interactive operator (#5854). Creating an edge with no -t/--type silently produces a blocks edge, which drops the dependent out of bd ready — the usual surprise when someone meant structural parent/child linkage. bd dep add now says so on stderr. Because blocks is the documented default and the majority-correct case, the note is gated: it is emitted only when stderr is a TTY, so scripted and agent callers (and CI, and bd dep add 2>log) never see it and are not trained to ignore stderr. --quiet silences it, as it does the other non-error stderr notices, and BD_NO_DEP_TYPE_WARNING=1 turns it off for operators who have internalised the default. --json output is unchanged.

  • bd show labels created_by as Created by:, not Owner: (be-ss66). The text view rendered created_by under a label naming a different concept. owner is a separate field holding a separate value — the git author email kept for CV attribution — exposed as owner in --json and never rendered in the text view at all. Two identities, one label, and the label named the one that was not on screen. This is a correctness fix and not polish: a reader asking "whose bead is this" reasonably reaches for what the label promises, and two defect analyses were filed off exactly that mistake in one evening, both measuring provenance from the wrong column and both retracted after another agent disproved them. bd show now prints Created by: <identity>. The --json payload is unchanged, and a repo-wide audit found no in-tree parser of the old Owner: text — the only other producer, bd github's repo-config display, names an actual repository owner and keeps its label.

  • beads_dir and repo_root are now OPTIONAL members of ContextResponse on GET /v0/beads/context. They are the only two members of the identity handshake that describe the SERVER's filesystem rather than the workspace's logical identity, and an absolute host path is not something a remote client can act on — it cannot open it. The document already called publishing them a cost "an operator accepts when binding beyond loopback"; while they were required, accepting it was the only conforming option. A deployment that does not want to disclose its filesystem layout can now withhold them and still serve a conforming body.

    No body changes. bd serve publishes both, always, exactly as before — the relaxation is a promise to clients, not a switch on the server. What changes is what a client may assume: it MUST tolerate their absence, and MUST identify a workspace by project_id, database and backend, which stay required. In Go, apigen.ContextResponse.BeadsDir and .RepoRoot become *string.

  • An explicitly configured Dolt server port now outranks the ambient BEADS_DOLT_SERVER_PORT/BEADS_DOLT_PORT environment variables (be-wf9a.1, be-9tju, GH#4052). ServerPort resolution now carries a source — which step of the chain produced the port — and the env read only applies when that source is not an assertion by the user or by tooling acting for them. Two deliberate behavior changes follow. First, a port set by the caller before the store opens (a library caller filling in Config.ServerPort, or bd init --server-port) is no longer silently replaced by whatever the surrounding shell exported; previously the env var won unconditionally. Second, the legacy BEADS_DOLT_PORT spelling no longer overrides a port resolved from .beads/config.yaml (dolt.port), Dolt's own config.yaml (listener.port), or metadata.json (dolt_server_port). The current BEADS_DOLT_SERVER_PORT spelling is unaffected — it sits first in the resolution chain and still wins over all of them.

    Compatibility: a workspace that relied on exporting BEADS_DOLT_PORT to redirect away from a port pinned in config now connects to the pinned port instead. Switch to BEADS_DOLT_SERVER_PORT, or change the pinned value. Ports that bd resolved for itself — the gitignored .beads/dolt-server.port, and the shared-server default 3308 — stay non-authoritative and remain overridable by either spelling, which is the case that kept working throughout.

    A related failure goes away with it: because bd's own bookkeeping is no longer mislabeled as user configuration, a stale .beads/dolt-server.port left by a crashed server no longer makes auto-start refuse to start with "bd will not silently use a different port than the one you configured" when nothing was configured. Auto-start warns and retargets, as it did before. Shared-server mode still fails closed there — a repo-local auto-start is a different database, not a port refresh — but now says so in its own words.

  • bd reclaim summarizes the leases its replica guard declined instead of naming every one, every run (wy-sp2l4). A lease granted by another replica is by construction never reclaimed here, so the audit was not a one-off: it repeated identically on every run, one stderr line per stranded remote lease. A supervisor on a 1-minute timer against a federated store with 47 of them printed 47 lines a minute, indefinitely. The default is now ONE line — reclaim: skipped 47 stale leases granted by other replicas — "mini" (30), "mini2" (17), not this node ("laptop")… — naming at most three replicas and collapsing the rest into a count, with the exact total preserved. bd -v (or BD_DEBUG=1) expands it to the per-lease lines, itself capped at 20 with a collapsed tail. Still stderr-only: bd reclaim --json owns stdout and its payload is unchanged. Under --quiet the audit now skips its queries entirely rather than running them to discard the output.

  • Actor matching decodes an exact -- run to / instead of collapsing it to a generic separator (be-p7dzx, bundled be-vc51). Identity comparison — used by ExpectedAssignee on release and update, and by claim/close/unclaim authorization — canonicalizes a run of ., _ or - to one separator so the same identity spelled under different layers' conventions compares equal. An exact two-byte -- run is now excluded from that collapse: it is gascity's session-name encoding of a rig-qualified agent's /, so it decodes to /. Compatibility: gastown--mayor now matches gastown/mayor, and stops matching gastown__mayor and gastown-mayor. That is the intended fix rather than a side effect — gastown--mayor is the agent mayor on rig gastown while gastown__mayor is the dotted alias gastown.mayor, so matching them was a widening — but any stored -- spelling that is NOT a gascity slash encoding changes equivalence class silently, with no error to notice. Longer or mixed runs, __ and --- included, are unaffected and still collapse.

  • bd ready --claim now refuses a row cap under --proxied-server. The proxied ready role cannot enforce --max-rows/BEADS_MAX_ROWS, so bd fails loudly rather than silently dropping the limit; --claim no longer exempts the command. Agent rigs that set the cap globally must unset it for proxied bd ready --claim. Direct mode is unchanged, and a claim there still succeeds against a ready pool larger than the cap. (#6269)

Fixed

  • Server-mode issue mutations no longer revert concurrent writers' committed rows (#5740). The issue-mutation path ran DOLT_ADD/DOLT_COMMIT inside its still-open SQL transaction; DOLT_ADD stages the whole table as of the transaction's BEGIN-time root, and the Dolt commit was minted before the commit-time merge that reconciles concurrent writers — so a mutation could silently write every row another session had committed during its window back to that session's BEGIN-time value, seen in production as claims reverting minutes after they were made. The Dolt commit now runs after the SQL transaction commits, against the post-merge working set. Its failure is logged rather than returned: the data is durably committed by then and rides the next Dolt commit, and treating an applied mutation as failed is what made retries double-apply and claim verification disown claims the caller held. Reported and fixed by @nova-submodules.

  • Removed-backend rejection now detects an existing Dolt database and gives the exact metadata.json heal (change "backend" to "dolt", or delete the field) instead of a destructive export/reinitialize path. Workspaces that bd v1.2.x opened as Dolt despite a stale "backend" value are one edit from healthy, and the message now says so. Rejection remains fail-closed and read-only — no storage database is opened or modified — and still exits 1. (Detection helper ported from @steveyegge's #4740.)

  • bd migrate no longer aborts with a duplicate-primary-key error when the same dependency edge exists in two typed target columns; duplicate edges are now merged deterministically (issue-ref wins over wisp/external), renaming an issue no longer leaves its dependency rows on a stale primary key, and databases left half-rekeyed by earlier binaries are detected and repaired on the next migration pass (#5268).

  • Legacy tracked ignored_schema_migrations self-heals by untracking it at open time, unwedging bd dolt pull on upgraded databases (#4356).

  • A v1.1.x/v1.2.x ignored-migration cursor is believed when leases.granted_node is merely not yet migrated, so upgrades apply only the pending ignored tail instead of replaying ignored/0007 and silently restamping wisps.updated_at (#5981 class, #5366 follow-up).

  • Auto-export heals after bd delete instead of permanently refusing — the orphan guard proves deletions against the store's own history (embedded and server modes), stores wedged by earlier binaries recover on first run, and export-state.json is now versioned (#5896).

  • bd flatten and bd compact now finish with a full Dolt GC (all generations), so re-running them on a previously-GC'd store actually reclaims the orphaned history and retires orphaned --as-of addresses; bd gc gains --full and hints when a default pass reclaims little (#5907).

  • --storage-class is honored on the proxied-server and --file routes and rejected as a plan-wide --graph flag; it was silently ignored there.

  • Delete and cascade dep_remove journal rows are attributed to the requesting actor (#5985); bd events export/tail now say when the journal is disabled.

  • Truthful errors when a server-scoped credential cannot reach the requested database (missing / denied / wrong-session attribution); --init-if-missing is honored in proxied-server mode.

  • The aux row re-key survives dolt#11131-class encoding drift instead of failing mid-upgrade with a raw invalid hash length error. Drifted tables are skipped with a warning, recorded clone-locally in aux_row_rekey_drifted, and re-keyed on a later pass, rather than leaving the database unopenable. This already reached 1.1.x and v1.2.2 users — it is the [1.1.2] fix — and lands on the main line here, so a store upgrading from v1.2.2 is not newly exposed (#5064, #4380). Diagnosis and fix by @marcodelpin, carried and reworked by @maphew.

  • Server mode honors Config.LenientOpen, so the dirty-table guard's documented recovery is executable. The refusal tells you to run bd dolt commit / bd vc commit, but against an external Dolt server those commands could not open the store to do it — the flag reached server mode and newServerMode never read it, while the field's own doc comment said "Ignored in server mode". That left the #4566 deadlock intact for exactly the case with no way out: embedded mode can move the database directory aside and bd bootstrap, and there is no equivalent against a shared server. Fixed by @Toady00 (#5783, #5781).

  • bd bootstrap probes git remotes for Dolt data instead of rejecting them outright (#6037, #5743, #5663). bd init deliberately persists a git-origin-derived sync.remote into the tracked .beads/config.yaml — a git origin is a valid Dolt remote before refs/dolt/data exists, because the first bd dolt push creates that ref on the same remote — but bootstrap classified those URLs as code-repo URLs and returned before the origin auto-detect that would have hydrated the clone, git+https:// and git+ssh:// included. Committed sync.remote values from earlier releases hydrate again. The failure was silent because one outcome code covered both "a database already exists, nothing to do" and "I refused and left you with nothing": bootstrap printed ✓ Database already exists, exited 0, and created nothing, after which bd init --reinit-local pointed back at bd bootstrap in a closed loop. Bootstrap now exits non-zero whenever it declines to set up a database.

  • --set-metadata again stores numbers/booleans/null as typed JSON scalars (v1.2.2 behavior); string-forcing lives on --metadata-json. Values written by in-window dev builds are left as strings; re-apply the flag to retype a key.

  • bd create --json (and POST /v0/beads/issues) again echo sub-second RFC3339Nano timestamps; reads (show/list) remain second-precision as in v1.2.2.

  • bd purge and bd prune select candidates by TIER, so a typed wisp is ephemeral however it was minted (#5995). The two sweeps split the world between them — bd purge clears the ephemeral tier, bd prune the durable one — but membership was decided by the raw ephemeral column, and a wisp minted before that column was set carried a wisp_type with ephemeral = 0. Such a bead belonged to neither sweep: bd purge reported "No closed ephemeral beads to purge" while bd prune skipped it as a wisp. One production database had 858 such rows / 7.3 MB that no sweep could ever reach. The tiers are now complementary predicates over both columns — ephemeral is ephemeral = 1 OR wisp_type is non-empty, durable is its exact complement — evaluated across both planes rather than taking the wisps-plane fast path, because legacy typed wisps live in the issues table. A non-empty wisp_type now also implies ephemeral at mint, on every create path, so no new bead can land in the gap. A previously unreachable typed wisp is now a bd purge candidate, so the first purge after upgrade may clear considerably more than usual. NoHistory beads (wisps plane, no wisp_type) stay durable-tier, unchanged.

  • bd prime says when it could NOT read the memory plane (#5877). A broken or unreachable store made prime omit the memory section entirely, so a session that woke with zero recall looked exactly like a workspace that simply has no memories — same output, same exit 0, no way for a fleet operator to tell them apart. Prime still never fails a session-start hook, but a memory read it cannot serve now renders Skipped: beads storage unavailable (<error>) — persistent memories were NOT injected this session, alongside the timeout banner that already existed for a deadline. Both the classic and the proxied-server route share the banner. A workspace-less directory, and a healthy store with no memories, stay silent as before.

  • bd no longer serializes every invocation behind the schema-init advisory lock. The schema-init probe ran entirely inside the database-scoped GET_LOCK('bd_schema_init:<db>', 5), so on a shared Dolt server every bd process queued behind every other one to discover that it had nothing to migrate. Measured on an 18-seat rig: 1500 is_free_lock samples over 14s found the lock held 96.7% of the time, held runs had a 673ms median and a 4.2s max, and free gaps had a 0ms median — the lock was handed straight from holder to holder. That queue was 0.4-2.4s of pure waiting on every claim, heartbeat, list, comment and mail check. MigrateUpWithLock now answers the steady-state question first, without the lock — it puts the session on the target database (proving the database exists before issuing any statement that could fail, so a fresh bootstrap still falls through to the locked CREATE DATABASE; callers whose pool already names the database inject nothing and the probe simply declines a session that is elsewhere), then asks whether both migration cursors are at this binary's latest with the content-hash column and custom-status/type backfills done, whether all the canonical dolt_ignore patterns are present, and finally whether the migration lock is free — and takes GET_LOCK only on the path that can actually migrate. That last term is what keeps the probe honest while a peer is mid-pass: the cursors land before the backfills and rekeys finish, so a held lock, not the cursors alone, is the proof that nobody is rewriting the tables underneath this caller. The probe issues no writes — its one session mutation is the USE that pins the target database — and fails closed: any error, any unreadable state, and anything short of provably converged falls through to the locked path unchanged, as does any caller carrying fresh-bootstrap heal authority.

  • Incremental auto-export now actually takes the incremental path (#5806). Change detection compared GetStateHash() values — a hash of the entire database plus working set (DOLT_HASHOF_DB()), not a commit — but ChangedIssueIDs feeds both endpoints straight into dolt_diff(), which only accepts real commit hashes or the literal WORKING. Every incremental attempt therefore failed to resolve a diff and silently fell back to a full export: correct output, but the incremental short-circuit this feature exists for never actually ran. The "to" endpoint is now WORKING (dolt_diff's own literal for the live working set) paired with the previous export's real commit hash as "from", so the incremental path resolves and fires as designed.

    Three format/scope regressions surfaced alongside the dead code path, since nothing had ever exercised it end-to-end: the incremental patch could leak memories into the auto-export output (full export filters them; the patch path did not), could include the configured owner's own issues where full export excludes them, and omitted the _type discriminator field full export always writes. All three are now pinned by tests that diff a patched run's output against a from-scratch full export of the same state. The patch write is also now atomic (write-temp + rename), matching the full-export path's existing guarantee.

    The owner and _type fixes match full export exactly. Memory records deliberately do not, and the difference is worth stating: where the full path refuses to overwrite an issues.jsonl that already contains memory lines (guardAutoExportOverwrite), the incremental path preserves them verbatim and proceeds. Preserving is the better behavior — a manually seeded export survives patching byte-for-byte instead of wedging — so the incremental path keeps it rather than adopting the refusal.

    Scope: this is a server-mode (DiffStore) improvement only. EmbeddedDoltStore implements neither DiffStore nor StateHasher, so in the default embedded mode the incremental path is inert (incremental skipped — store does not implement DiffStore) and every cycle is still a full export. The deletion-proof guard below started out server-mode only for the same reason; #5896 is now closed for embedded mode too — see the store-history heal entry below.

    Caveat: the diff anchor only advances on real commits. In server mode dolt auto-commit is off, so with no commits being made the anchor stays put and the diff range grows with every cycle. That is correct but not free: once the change set crosses the 5000-id threshold, every export falls back to a full rewrite until something commits and the anchor moves forward.

  • bd delete no longer wedges auto-export in server mode (#5806, part of #5896). Auto-export's orphan guard refuses to overwrite an issues.jsonl holding issue records absent from the local store — the #4988 protection against a JSONL that has run ahead of the database. A normal bd delete produced exactly that shape, so a single delete left issues.jsonl permanently stale, with a "refusing to overwrite" warning on every later command. The guard now asks dolt_diff whether each JSONL-only id was genuinely removed since the last export's anchor and, when proven, stops treating it as corruption.

    The proof is deliberately narrow. It requires diff_type='removed' on the issues table itself, so a label-, dependency- or comment-only change, a delete-then-recreate, a wisp, or any diff error all still refuse. It also requires the anchor to still be reachable from HEAD: after a DOLT_RESET --hard, a branch checkout, or any other data-dir rewind, a row reads as "removed" without anyone having deleted it, and honoring that would drop a live record. Both conditions must hold, or the guard refuses as before.

  • Disabling telemetry no longer strands the queued eventsData backlog forever (GH#5712). bd send-metrics early-returned on disabled metrics before its prune step, and the spawn gate refused to schedule the child at all when disabled — so the one machine whose queue can never again drain by upload (the one that just opted out) kept every queued batch and orphaned emitter temp indefinitely (2M+ files / 15.8GB observed on one control VM). With metrics disabled, bd now still spawns the detached child under the same marker-first throttle, and the child prunes by the normal TTL/cap policy and exits without POSTing anything. Until that pre-existing backlog ages out, a host that just opted out while its queue still holds batches younger than the 7-day TTL (and under the file/size caps) keeps forking the throttled, detached, network-free prune child every flush interval — reclaiming nothing until those batches age past the TTL — so new prune-only bd send-metrics processes on a freshly disabled host are expected during that bounded window and never POST. Once the backlog has decayed empty, spawns stop entirely; a machine that never enabled telemetry has no queue directory and never forks.

  • bd show on a deleted or purged issue no longer prints text identical to an ID that never existed (ga-m6inyb). bd delete and wisp purge both remove a row (and its events) from the live tables with no trace left behind, so "not found" was collapsing two states that call for opposite responses — archived (success) and never-existed (possibly lost data) — into one observable. This had a real cost: a gate-reviewer session showed a legitimate issue successfully, hit an unrelated mysql i/o timeout on a retry moments later, got the identical "not found" text for the same ID, and reasonably concluded data had been lost — when the issue had simply been archived in between. bd show on a missing ID now adds a second line — Hint: this ID may have never existed, or may reference a deleted/purged record with no trace left in the live database — try 'bd history <id>' — pointing at the existing, opt-in bd history <id> for anyone who wants to check further on a non-wisp issue (wisps are never committed to history, purged or not, so there is no signal left to recover for them). JSON output gains an additive hint field; the existing error text is unchanged. Direct-mode text output changes for one path: the single-ID not-found branch's first line changes from Error fetching <id>: no issue found matching "<id>" to Issue <id> not found, matching the wording the multi-ID and proxied-mode paths already used (those two are unchanged aside from gaining the new second Hint: line).

  • Explicit commits now sweep the entire working set and report no-ops honestly. In server mode, bd vc commit and bd dolt commit previously routed through the config-excluding automatic-commit path, so out-of-band config changes could remain dirty even though the command exited successfully. Both commands now include config while continuing to exclude ignored wisp tables. A clean or otherwise non-committable working set reports Nothing to commit (committed: false in JSON) instead of attributing an unchanged HEAD to a new commit.

  • Every CLI label write now normalizes its input, as every label filter already did (#5813, fixes #5812). utils.NormalizeLabels (trim, drop empty, dedupe) was applied to list, search, ready, orphans, count and workapi — and to no write path at all. The filter trimmed its input and the store did not, so a label could not match its own filter: bd create --labels 'theme:a, theme:b' stored theme:b with pflag's leading space intact, and bd list --label theme:b then returned nothing. The failure is silent in the worst way — a filtered list quietly short of rows is indistinguishable from a complete one. One real database had 150 such rows across 111 issues, found months later only because theme-by-theme counts refused to reconcile. create, quick, update (--add-label, --remove-label, --set-labels) and tag now normalize on the way in. --remove-label matters as much as the additive flags: an untrimmed ' theme:a' silently removed nothing.

    bd tag is included deliberately. Its help calls it "Shorthand for bd update <id> --add-label <label>", and it was the last CLI door still storing a positional verbatim. Normalization happens before the direct/proxied route split, so the two routes cannot disagree. A label that is only whitespace is now refused rather than dropped: the plural flags can discard an empty element and still honor the rest of the request, but bd tag has one label to add, so dropping it would report success having written nothing.

    Import and batch ingest are deliberately excluded and keep round-trip fidelity: what was exported is what is restored.

  • bd show no longer corrupts shell globs and other quoted wildcards in a description or notes (#5799). CommonMark lets a * or _ run act as both an opener and a closer when the characters flanking it fall in the same class, and a quoted glob is exactly that shape — the * in '*.captured' sits between ' and .. Two such runs therefore paired, including across lines, since goldmark joins a paragraph's lines before parsing. Everything between them rendered as emphasis: with color the asterisks were consumed and silently vanished from the command, leaving a still-plausible but different one on screen; without color, glamour's notty style wrote a literal ** into the middle of the text. Storage was never affected — bd show --json always returned the body byte-for-byte — but the rendered text is what a reader acts on, and what an agent reads when it takes its instructions from a bead. Runs that can act as both opener and closer are now hidden from the parser and restored afterwards, so bd show prints what was stored.

    The known trade, in two narrow shapes, both pinned by test: an authored closer that is itself both-flanking no longer pairs, so see *"quoted"*, next renders literally rather than italicized (color path only — notty already printed it literally); and a backslash escape reaching for a both-flanking delimiter keeps its backslash, so \*.captured renders as \*.captured. An escape whose delimiter is not both-flanking, such as a \*literal\* b, is unchanged. Both are shapes CommonMark itself treats as ambiguous, and for a bead body — stored plain text, not authored markdown — showing the stored characters is the better of the two failures. Ordinary emphasis (**bold**, *italic*), SELECT * FROM t, code spans and fenced blocks all render exactly as before.

  • bd blocked silently ignored the label filters it already accepted (#5822). types.WorkFilter has carried Labels, LabelsAny and ExcludeLabels all along, and bd blocked built one, but GetBlockedIssuesInTx only ever read ParentID. Programmatic callers that set them got unfiltered results back and no error — a full list where a scoped one was asked for.

  • Proxied-server mode (and bd serve) ran schema migrations on every store open without consulting any migration gate, so an upgraded client silently migrated a shared database — including from read-only commands. The gate now runs on that path too: read commands warn and continue on the current schema, writes refuse, and bd serve fails to start until the schema is reconciled. (#5043)

  • bd migrate schema now works in proxied-server mode instead of refusing. The migration runs during the provider open under the verb's consent. (#5043)

  • Migration-gate and schema-skew failures on the proxied open path and at bd serve startup now print their full actionable block instead of a single truncated line. (#5043)

  • bd doctor now honors strict --readonly and the migration freeze (#6028). bd doctor --fix / --clean refuse to mutate under either gate — the freeze refusal exits 14 and names the marker, and there is no doctor-side override — and plain bd doctor diagnosis no longer rewrites .local_version or auto-applies schema migrations while a freeze is active or --readonly is set; it reports the pending migration instead. Both gates key on the directory doctor was pointed at, not just the one it was launched in. Diagnosis itself keeps working under a freeze.

  • An externally-managed Dolt sql-server is no longer mistaken for bd's own, which had silently disarmed the shared-store migrate gate (#6118). When the port arrived from BEADS_DOLT_SERVER_PORT, from bd init --server-port, or from a stale port file left by a server that had since died, bd classified a genuinely shared server as workspace-owned and migrated its schema in place — no prompt, no warning, exit 0 — after which every older client on that server was hard-refused with schema version mismatch. Ownership is now proven from bd's own live port and PID record rather than inferred from how the endpoint was named: bd must have bound that port and the process answering it must still be alive. Everything else is shared, and stays gated.

  • A dolt.mode: server workspace declared in .beads/config.yaml is no longer refused as a legacy workspace (#6119). The upgrade guard resolved the connection mode from metadata.json alone, so a workspace that names its mode only in config.yaml fell through to the embedded arm and was refused outright — including workspaces this release had just created and was already operating on. The guard now resolves the mode through the same IsDoltServerMode precedence chain the rest of bd uses, and a .local_version witness naming bd 1.0 or later vetoes the legacy verdict in every mode rather than only in server mode.

  • A server-mode workspace with no metadata.json no longer opens a phantom embedded database and answers bd list with a false-empty result and exit 0 (#6120). Config substitution now gates on IsDoltServerMode, and BEADS_DOLT_SERVER_MODE is honored on that path instead of being ignored.

Security

  • Release binaries now build with Go 1.26.7 instead of 1.26.5, closing seven standard-library advisories reachable from bd: quadratic net/url path resolution, html/template JavaScript-regexp context tracking, unbounded post-handshake crypto/tls messages, a missing ReadHeaderTimeout on net/http's unencrypted HTTP/2 check, unbounded recursion in encoding/xml and encoding/asn1, and the x/net/idna Punycode bug in net/http's vendored copy. bd serve and bd's outbound TLS calls exercise these paths. The go directive stays at 1.26.5, so nothing changes for code importing the beads module.

  • Dependency bumps clearing all 25 open advisories. golang.org/x/crypto 0.55.0, golang.org/x/mod 0.40.0, klauspost/compress 1.18.7, moby/go-archive 0.3.3, kin-openapi 0.144.0 with oapi-codegen 2.7.1, and a refreshed integrations/beads-mcp/uv.lock. None of the fixed code was reachable from the shipped bd binary: the go-archive and x/crypto/ssh paths are testcontainers-only, the compress bug is in the s2 codec bd does not link, and the kin-openapi critical is in openapi3filter middleware that enters the graph only through the tool directive for spec codegen. No behavior change; no dolt bump was required.

Documentation

  • Known limitation: multi-rig prefix routing (routes.jsonl) is not supported with proxied-server rigs — cross-rig bead gates stay pending, bd close of a bead gate in proxied-server mode requires bd gate check or --force, and routed lookups cannot open proxied-server targets (#5861).

  • The heartbeat/re-home invariant and the two states it can strand are now documented where the code lives (wy-sp2l4): a heartbeat proves the holder is alive, not that the lease moved, so granted_node is backfilled but never overwritten. A renamed replica (mini → mini2) and an imported foreign lease whose holder name also exists locally therefore keep reading foreign while a local heartbeat keeps them alive, recoverable only with bd reclaim --any-replica. Reviewed and KEPT rather than re-homed: re-homing would let a heartbeat silently reassign enforcement of a lease another replica granted — the robbery the guard exists to prevent. Recorded in HeartbeatIssueInTx, docs/multi-agent/federation.md, and bd reclaim --help.

[1.2.2] - 2026-08-15

Recovery release, cut from the v1.1.2 tree on branch release/v1.2.2 (not from main). v1.2.0 and v1.2.1 were published by accident on 2026-08-11 without release testing (v1.2.0's tag burned pre-publish; only v1.2.1 reached users). This release supersedes them by re-releasing the tested 1.1 line: it is the v1.1.2 code under a higher version number, so every install channel (Homebrew, npm, the install script, go install ...@latest) moves forward onto tested code. The 1.2.x-only features listed under [1.2.1] below are not in this release; they remain on main for a properly tested future release.

If v1.2.1 migrated your database (running any command once was enough), v1.2.2 stops with schema version mismatch: database is at v65, binary knows up to v53. See the recovery guide (site, repo copy at the tag) for the two-minute fix (roll the schema cursor back to v53) and a verified stopgap (BD_IGNORE_SCHEMA_SKEW=1).

Changed

  • The forward-skew error recognizes the accidental-release window (on the release branch): a v53 binary opening a database at schema v54–v65 points at the recovery guide instead of advising "install the latest release" (which would loop back to that binary).
  • go.mod retracts v1.2.1, v1.2.0, and v1.1.1 so go install github.com/steveyegge/beads/cmd/bd@latest resolves to this release instead of the accidental one. (The retract block is also carried on main so future tags keep the retractions.)

[1.2.1] - 2026-08-11

(Released as 1.2.1: the v1.2.0 tag burned pre-publish on a freebsd cross-compile failure in the release build — fixed below — and a burned tag is never reused, per the v1.1.1 precedent.)

Added

  • --brief on bd list and bd ready, and brief on the two HTTP listings (#5546, #5554, #5586). The listing commands return the whole body of every row they page. On one real 16,051-issue store bd ready --json returned 454,021 bytes with notes plus description accounting for 76% of it, on exactly the call an agent makes to decide what to work on next. --brief omits the free-form text (description, design, acceptance_criteria, notes, payload, waiters), measuring 93.4% smaller. Filtering is untouched: --desc-contains and its siblings select rows, and this selects fields. It is opt-in and the default payload is unchanged.

    The response carries no marker for the omission. An omitted field is indistinguishable from a genuinely empty one, so only the caller that passed the flag knows its rows are partial; there is no is_lite_partial in the JSON. The one visible marker is in text output, where bd list --long --brief prints Description: (omitted by --brief) rather than an empty section.

    The flag is refused wherever it could not be honored, rather than accepted and dropped. On bd ready it requires --json and is refused with --claim, --gated, --mol and --explain: the projection reaches the driver through the counts query alone, which those routes do not run. On bd list it works in text mode too, and is refused with --watch, with the --parent tree walk, and with --format, where a caller's template could print a dropped field with nothing to mark it.

  • --brief-deps on bd show, and brief_deps on the HTTP detail read (#5546, #5547, #5549). bd show --json inlines every dependency at full body, so reading one issue costs the free-form text of everything it depends on. On one real 16,051-issue store a single bd show --json returned 214,456 bytes, of which 193,039 was the dependencies array and one dependency carried a 180,975-byte notes field. The dependents side of the same detail view already had a shallow shape for this reason (be-4d36f2, where hub beads cost 5-13 GB); dependencies never got it. The new flag projects each dependency down to id, title, status, issue_type, priority and dependency_type, which measured 89.3% smaller on that call. It is opt-in and the default payload is unchanged.

  • bd serve gets optional bearer authentication, and requires it beyond loopback (#5516). The help used to say "no authentication and no TLS", and the first half of that is what kept the server on a developer's loopback. Two flags answer it, and both default to today's behavior: a bd serve with no arguments is byte for byte the server it has always been.

    --auth-token-file names a file of accepted tokens, one per line, all accepted. Every operation except GET /healthz then requires Authorization: Bearer <token> — GET /v0/beads/context included, because it reports the repo root, beads directory and database name. There is deliberately no --auth-token: a credential passed as an argument is readable out of ps by every local user. BEADS_SERVE_TOKEN_FILE is the environment fallback.

    Rotation is a file rewrite, with no restart. The accepted set is not a startup snapshot: the file is re-read while the server runs, at most once a second for the whole process, on the accepting path as well as on a mismatch — which is what makes REVOCATION work and not just rotation. Write {new,old}, roll the clients, drop old; both take effect within about a second. A failed or empty re-read keeps the last-good set and logs, so a writer that truncates before writing cannot lock every client out.

    --allow-non-loopback now requires a token file. Beyond loopback, reaching the address would otherwise be the whole authorization. --insecure-no-auth is the explicit, auditable way to say you meant that anyway. There is still NO TLS: a deployment beyond loopback supplies confidentiality itself. --allowed-host (repeatable, exact, no wildcards) adds Host header values to the DNS-rebinding allowlist, which is what a service deployment behind a DNS name trips over first.

    401 / unauthenticated joins the frozen code vocabulary, with WWW-Authenticate: Bearer and a FIXED detail that never echoes the presented credential. A missing header, a wrong scheme and an unrecognized token are one code deliberately — telling them apart would tell an unauthenticated caller which guess was closer. A token is a shared secret granting the whole surface, not an identity, so actor is still caller-asserted provenance and not the authenticated principal.

  • Six more v0 operations, and the guarded-write token they compose from (#5506, #5507, #5508, #5510, #5535, #5536, #5540). Each is the wire half of a role that already existed and was already pinned on three legs.

    • POST /v0/beads/issues/{id}:release — give a claim back, what bd unclaim spells. It is NOT idempotent, and that is the one thing to read before adopting it: a release over a row holding no claim is 409 / not_releasable, never a 200, because the post-state is anonymous. Do not read the code as "already released" — it also covers "the status will not accept a release", which is true of a row that is still ASSIGNED. Read the row. expected_assignee is a compare-and-set on the holder and force ignores it; the two may not be sent together.
    • POST /v0/beads/issues:claimNext and POST /v0/beads/issues:batchClose — take the next ready bead in one round trip, and close many as one act.
    • GET /v0/beads/issues:count and GET /v0/beads/dependencies:count — the two counting reads, each with its role's whole filter surface. dependencies:count bounds issue_id at 100 anchors.
    • GET /v0/beads/issues/{id}/related — one anchor's neighbors, direction- parameterized and type-filterable. The length of items is a NEIGHBOR count and not an EDGE count: a target that is an external: reference or another repository's id has no far end here and is silently not a neighbor, while the same edge IS a row on GET /v0/beads/dependencies.
    • expected_version on close, reopen and delete, and revision on GET /v0/beads/issues/{id} — the row's optimistic-concurrency token, so a read-modify-write loop can start from a READ instead of seeding its first guard from a write it did not want to make. A miss is 409 / precondition_failed and writes nothing. revision is deliberately NOT on the listing rows: types.IssueWithCounts is also the JSONL interchange record, so a token there would write a per-write-random value into every bd export line and every auto-flushed issues.jsonl.

    not_releasable joins the frozen 409 vocabulary. Every one of these carries the claim's posture unchanged: actor is caller-asserted provenance, hooks do not fire, the per-command auto-commit machinery does not run, and the only durable effect is the single storage commit the role makes in its own transaction.

  • bd serve answers a Bd-Project-Id header, and says so in capabilities. A client that knows which workspace it means to address stamps the request with that workspace's project id; a server that serves a different one refuses before any database work, so a misdirected read or write mutates nothing. An ABSENT header is served exactly as before, so this is additive wire surface and not a new precondition. GET /healthz and GET /v0/beads/context are exempt — liveness must answer whatever workspace a caller believed it reached, and the handshake is where a client LEARNS the id to stamp with. Detect it with the project.enforce capability.

  • A durable events journal, and bd events to read it (bd-opisf). External tooling that wants to stay in step with a workspace had two options and neither was a feed: a fire-and-forget script hook that may not run, or polling whole snapshots and diffing them — which still misses anything that changed twice between two reads. The journal is the third. Every committed issue mutation writes one ordered record, in the same transaction as the mutation itself, and a consumer replays those records from a checkpoint it can resume. Each record carries the operation, the mutated id, and the issue's full post-mutation snapshot (including is_blocked), so a mirror stays correct without re-querying the graph.

    It is off by default and opt-in per workspace: bd config set events-journal true, or BD_EVENTS_JOURNAL=1. Nothing is recorded while it is off, so a workspace that never enables it pays nothing.

    On means bounded. Two retention floors define the window every consumer is guaranteed — events-journal-retain-days (default 7) and events-journal-retain-rows (default 100000), each disabled by setting it to 0 — and beads enforces them for you: after a mutating command commits, and on a timer inside bd serve, it deletes the prefix the floors do not protect. The pass is throttled by a persisted watermark (about one an hour per workspace — or sooner after a large burst of writes), runs in its own transactions outside the mutation's, is capped at a few batches so a long backlog drains over several commands rather than stalling one, and can never fail a command — a failure is logged and skipped. It maintains the workspace whose command triggered it, so a workspace only ever written remotely (a routed bd create --repo) relies on commands run in it, or on its own bd serve. Setting both floors to 0 keeps every record forever; events-journal-auto-prune false keeps the floors but leaves deletion to you. All four are config.yaml keys with BD_EVENTS_JOURNAL* environment equivalents.

    bd events tail --since <seq> prints records as JSON lines and --follow keeps printing them as writes commit; bd events export prints the journal from the beginning; bd events prune --before <seq> takes an earlier, on-demand cut below the floors — it cannot cut deeper than they allow, so shrinking the retained window means lowering them. The floors are a recent-window guarantee, not a consumer watermark: size them for the longest outage a consumer must survive.

    A read that cannot resume fails instead of lying. When --since falls below the oldest retained record, the read neither skips ahead to the surviving suffix nor returns an empty success: both are silent record loss that a cursor cannot detect. It exits 1 with a typed events_journal_truncated error carrying since, floor and head, so a consumer can choose between resuming at floor - 1 with a known gap and re-baselining. Under --json that is the error payload; mid---follow it is one compact JSON object on a line of the JSONL stream it interrupts, because the consumer on the other end is a line reader. An interior gap — which nothing in bd can produce, since a prune only ever removes a prefix — refuses the same way.

    The journal is clone-local working-set state (dolt_ignored): never versioned, never pushed or federated, per branch, and per replica — each clone counts its own seq space, so a checkpoint from one replica is meaningless against another. bd dolt pull and merge-settled changes, raw bd sql DML, store-open migrations, and compaction rewrites are not journaled. The record contract and every boundary are documented in docs/reference/events-journal.md.

  • is_blocked is a documented optional member of the /v0 OpenAPI issue schemas (bd-opisf) — Issue, IssueWithCounts, IssueDetails, IssueWithDependencyMetadata and TreeNode, the five welded to the canonical Go struct. It is the persisted readiness projection: true when an open blocking dependency keeps an issue out of the ready set, derived from the dependency graph and maintained by the server, never set by a client. It is omitted when false, so absence means false. Journal snapshots are what set it; readiness itself is computed exactly as before.

  • bd serve grows the write half of the agent loop (v0 wire surface, #5410, #5417, #5422, #5423, #5429). v0 shipped with one mutation, the claim, which meant an HTTP client could take work and had to fork a bd subprocess to finish it. Five operations close that loop:

    • POST /v0/beads/issues/{id}:close and POST /v0/beads/issues/{id}:reopen — the two halves of the lifecycle move. Both are idempotent, and both say so in the body rather than in a status: a close of a closed issue answers 200 with already_closed: true, a reopen of an open one already_open: true, and each still carries the row.
    • PATCH /v0/beads/issues/{id} — partial update of one issue. The patch object publishes nineteen members; four of them (estimated_minutes, external_ref, due_at, defer_until) are nullable, and an explicit null CLEARS them. That set is closed and machine-checked against the document, so a null on any other member is a 400 rather than an unannounced clear. A patch that changes nothing answers 200 with changed: false.
    • POST /v0/beads/dependencies:add — up to 100 edges per request, asserted all or none, echoed back in request order. A refusal names the offending edge as edges[i].member, so a client learns which edge and which member rather than that something in the batch was wrong.
    • POST /v0/beads/dependencies:remove — removes exactly the edge the body names. Idempotent at the role: an edge that is not there is removed: false with a 200, never a 404, so a replayed teardown does not have to classify an error to discover it already ran.

    Every one of them carries the claim's posture, unchanged. The actor is caller-asserted provenance for the audit trail and not authenticated identity; hooks do not fire; the per-command auto-commit machinery does not run. The only durable effect is the single storage commit the role makes inside its own transaction, so durability is per request on every write, not just the claim.

    Three codes join the frozen error vocabulary, all 409 and all typed so a client never has to substring-match a message. not_closable is close policy refusing an unforced close of an issue with open children or a live blocker. dependency_exists is the pair that already carries an edge of a DIFFERENT type, and carries existing_type and requested_type. dependency_cycle covers both never-makes-progress refusals: the plain scheduling cycle, and a blocking edge against the issue's own ancestor or descendant — the second carries issue_id, blocker_id and blocker_is_ancestor, and their ABSENCE is what tells a client which of the two it got. Every extension member is read off the refusing transaction's own typed error; the hierarchy members have no other source, since the conflicting edge may exist only inside the batch that was rolled back. An edge naming an endpoint this database can see the absence of is a 400 on edges[i].issue_id or edges[i].depends_on_id, not a 404: the request BODY is what is wrong, and there is no path id to have missed.

    GET /v0/beads/issues/{id} takes two new parameters. include_comments populates comments with the full comment bodies and include_dependents populates dependents with the issues that depend on this one, each carrying its edge type. Both default to false, because both are extra reads nobody should pay for by accident — and when a caller does not ask, comment_count, dependent_count and comments_omitted still report what was left out, so an absent comments is never ambiguous between "no comments" and "not included in this response". A malformed boolean is this operation's own 400 invalid_argument / reason: "invalid_value", which is deliberately not the unknown_parameter a client should read as version skew.

    GET /v0/beads/context advertises issues.close, issues.reopen, issues.update, dependencies.add and dependencies.remove alongside the rest — derived from the registered handlers as always, so probing capabilities remains the way to ask one server what it can do.

  • The two dependency-endpoint refusals now carry an identity (bd-yby99.9). DependencyEditor.AddDependencies refuses an edge whose SOURCE names no row, and one whose TARGET names no row this database can see the absence of. Both were "an error" and nothing more — not the same error text on every backend, and impossible to tell apart from an infrastructure failure — which left the role's typed-refusal vocabulary with a hole every other refusal it can raise had already filled. They are now errors.Is-matchable as ErrDependencySourceNotFound and ErrDependencyTargetNotFound, carried by a *DependencyEndpointNotFoundError naming the refused edge and which of its endpoints was absent, all three re-exported from the root package beside the conflict types. What is NOT refused is unchanged: an external: reference and another repository's id are still accepted as external targets.

    The user-facing message is byte-identical on the embedded and store paths (issue <id> not found), and the proxied-server path now renders that message too instead of a raw foreign-key violation — the endpoint is classified from the refusing transaction rather than parsed out of the driver's prose, so the two plumbings agree on the text as well as the type. The shared DependencyEditor contract asserts the sentinel and the typed fields for each endpoint, alone in a request and mid-batch, with the mid-batch half reading the graph back at zero edges. RemoveDependency is unaffected: a removal that finds no edge is a success, not a refusal.

Changed

  • Text-input commands now REFUSE two sources instead of silently picking one (#5332). bd comment, bd note and bd comments add used to apply a precedence order when a caller gave both positional text and --stdin or --file: the flag won and the positional text was dropped with no message. Naming two sources is now an error — cannot combine positional text "..." with --stdin — because the dropped half was, in every case we could find, the text the caller actually meant. Single-source invocations are unchanged. Scripts that relied on the old precedence (passing both and expecting the flag to win) must drop the redundant argument.

    bd human respond gains the same layer, so its response text can come from positional args, --response, --file or --stdin; --response is no longer a required flag. bd human dismiss takes its reason positionally as well as via --reason. Both now accept trailing positional text (<id> [text...]), matching bd comment's contract — a second argument is free text, never a second issue id.

  • bd list --status=all no longer hides boolean-pinned beads (#5332). all promises every status, but the pinned exclusion was decided by comparing the raw selector string to pinned/hooked, so all fell through and kept forcing Pinned=false. The same string comparison also missed every multi-status set containing them, so --status=pinned,closed excluded the pinned beads it explicitly asked for. Both now lift the exclusion. Scripts that used --status=all as a way to list everything EXCEPT pinned beads will see pinned beads appear; add --no-pinned to keep the old result. --ready is unaffected: it forces status open and ignores the selector, so an all it never applies has no pinned side effect either.

  • bd human list matches bd list's status handling, and hides no bead type (#5332). It used to pass --status through unvalidated and show closed beads by default. Now done/frozen statuses and pinned beads are hidden by default, --status is validated against the built-in and custom status names (comma-separated sets and all supported, so a typo is an error rather than an empty list), and every bead TYPE still shows — gates, wisps, infra beads and templates — because a human label is an explicit request for a person's attention. bd human stats counts across every status, as before.

  • bd human stats classifies dismissals by close-reason PREFIX (#5332). It matched the substring dismiss case-insensitively anywhere in the close reason; it now requires the Dismissed prefix that bd human dismiss writes, shared as one constant so the writer and the classifier cannot drift. Beads closed before this with a lowercase or mid-string "dismiss" in their reason move from the Dismissed count to the Responded count. Forward behavior is unchanged.

  • bd label add/remove over many issues is no longer one atomic transaction (ga-26w10, #5489). Both routes — the direct one and the proxied-server one — now end at issueops.Lifecycle and issueops.Reader, which deletes the four-way (add|remove) × (wisp|durable) switch the proxied route hand-rolled below the roles and the second implementation of the command that came with it. The plane is resolved inside the role's own transaction, so the wisp branch is gone rather than moved.

    The deliberate delta: a multi-issue edit is N guarded updates rather than one raw transaction, so bd label add a b c mylabel records three history entries instead of one, and a failure on the third leaves the first two written where the old shape rolled them back. Neither is obviously right — the new one attributes each edit to its own issue and lands what it could — and the atomic shape is expressible today as one issueops.BatchApplier update item per issue. The proxied route's resolution failure also lost its label added: prefix, which named a write the command never attempted; both routes now print the direct route's message.

    bd label list-all and bd label propagate are unchanged and stay off the role: no role answers "every distinct label in this workspace", and propagate is a search plus a fan-out needing the same batch question.

  • The settings plane no longer serves the KV plane by any read (bd-rfwtv, bd-klko9). bd kv keys and the bd remember memories nested under them are user data stored as config rows; they ride in the settings table without being settings. Listing them was already excluded, but a caller naming an exact key still got the value — and since bd remember derives its key from the content it stores, the keys are guessable, so bd config get kv.memory.<slug> and GET /v0/beads/config/kv.memory.<slug> (a surface whose only credential is an optional shared bearer, and whose redaction decides on the key NAME while a memory's secret is in the VALUE) walked around the exclusion.

    A point read of a kv. key now answers exactly as a key nothing ever stored does — the echoed key, an empty value, a nil error — so a caller cannot tell a refusal from an absence. bd config show no longer prints those rows under source=database either; it reads the table raw and prints values in full, so it had inherited neither filter.

    Nothing is deleted and nothing becomes unreachable. The rows are untouched, and the surfaces that own them — bd kv get, bd kv list, bd recall, bd memories — read the store directly and are unaffected. bd config set and bd config unset still take a verbatim kv. key, which keeps the escape hatch for a wedged memory. What is gone is reading memory contents through bd config get, bd config list and bd config show.

    This is plane hygiene, not a confidentiality boundary. bd serve has no authentication, and it serves every memory in full through /v0/beads/memories by design — anyone who can reach the server can still read them. What this closes is the settings plane's habit of handing user data to callers that asked for configuration: the routes that get republished into agent transcripts and pasted into issues. Do not read it as a fix that makes memories secret from someone who can already reach the port.

  • bd search now includes closed issues by default (bd-t5yex). The dominant real-world search query is "was this already found/filed/fixed?" — exactly the query where silently excluding closed issues produced a false "no" (downstream repos grew shell wrappers purely to undo the old default). bd list keeps its open-only default. Narrow explicitly with --status open (or any status list) where the old behavior is wanted; --status all still works and is now redundant. Applies to both the embedded and proxied-server paths.

    Two consequences to know: matches beyond --limit (default 50) are dropped under a status-blind sort (priority, then created), so a broad query on a large DB can now fill the page with closed matches and silently drop open ones — use --status open or raise --limit when hunting live work. And bd query (plus the HTTP q= endpoint) deliberately keeps its closed- exclusion default with opt-in --all; only bd search changed.

Removed

  • BREAKING (published backend package): orphan handling is gone (bd-gwryr). BatchCreateOptions.OrphanHandling and its four modes never did anything except cost a query. Removed symbols: backend.OrphanHandling, backend.OrphanAllow, backend.OrphanResurrect, backend.OrphanSkip, backend.OrphanStrict, and the storage originals they aliased (storage.OrphanHandling and the same four constants), the BatchCreateOptions.OrphanHandling field, and issueops.CheckOrphan.

    Migration for out-of-tree consumers: delete the option. Every call site in bd passed OrphanAllow (or left the field at its zero value, which took the same branch), and OrphanAllow is exactly what the code still does — a hierarchical create whose parent is missing is accepted, and the child's auxiliary counter row is skipped because it would have no owner. Only OrphanStrict (reject the create) and OrphanSkip (silently drop it) behaved differently, and nothing but one conformance case ever passed them; a consumer that relied on either must now check for the parent itself before calling. OrphanResurrect was never implemented — it shared a branch with OrphanAllow, and the resurrect-a-deleted-parent behavior its doc comment described left with the JSONL sync layer long ago.

    Every hierarchical-ID create paid a SELECT COUNT(*) parent-existence probe to feed a switch whose only reachable arm ignored the answer. That query is now gone from the create path.

Fixed

  • --label-any is no longer silently dropped by bd ready and bd ready --claim. The ready-work WHERE builder emitted clauses for --label and --exclude-label but none for --label-any, so the OR-set filter was ignored on the ready/claim path (with or without --parent) on every backend — while bd list/bd search honored it. On an atomic claim this was dangerous rather than merely wrong: a worker fencing itself to its own lane (bd ready --claim --label-any lane-a --parent epic-1) would happily claim another lane's issue and believe it was fenced. --label-any now emits an OR-set membership clause that AND-combines with --label, --exclude-label, and --parent, exactly as the flag help promises; an exhausted lane now claims nothing instead of falling back to unfenced work.

  • FreeBSD builds compile again (#5661). The dbproxy process-identity arc (procid, unverified-process) shipped linux/darwin/windows implementations with no fallback, breaking GOOS=freebsd compilation — caught only by the release build's cross-compile (GH#5662 tracks the CI gap). Unsupported platforms now get stubs that fail proxied-mode spawns with a clean, actionable error; classic (non-proxied) bd is unaffected, matching what v1.1.2 shipped on freebsd.

  • Telemetry no longer taxes every bd invocation. Two startup costs paid on every command, measured during the post-wave startup audit, are gone. First, the machine-scoped distinct ID was recomputed on every invocation — a fork of the platform machine-id probe (ioreg on macOS, 20.2±1.2ms) — even with BD_DISABLE_METRICS=1 and even for bd --version. The ID is now resolved only when metrics are enabled and cached at ~/.beads/machine-id (0600), so the probe runs at most once per machine; a probe failure is retried next run rather than cached. Second, every invocation unconditionally spawned a detached bd send-metrics child — a full re-exec of the binary plus an HTTPS upload attempt, with no check that anything was queued. The spawn now requires at least one queued event batch and is throttled to one attempt per 5 minutes (marker: eventsData/.last-flush); queued events still upload, just batched. Telemetry content, opt-out semantics, and the sanctioned endpoint pinning are all unchanged.

  • A long or multi-paragraph close reason renders as body text in bd show (#5595). Every other free-text field — description, design, notes, acceptance criteria, comments — reaches the terminal through the markdown renderer, which word wraps and indents. The close reason did not: it was formatted into the metadata block beside Owner: and Created:, so it never wrapped and its second and later lines read as separate metadata entries, a blank line and a bare - bullet sitting directly under Created:. bd close --reason-file exists so agents can write structured close reports, and those were exactly the reasons that came out corrupted. A reason that still fits one metadata line — nearly all of them, including one-liners that arrive from a file with a trailing newline — is unchanged; anything larger now gets a CLOSE REASON section rendered by the same call the other body fields make. The JSON payload is untouched. The compaction savings line moved into the metadata block at the same time, so it can no longer be stranded below that section, and all five bd show render paths now report it alike.

  • Script hooks now fire on both write plumbings (bd-opisf). bd has two write plumbings and only one of them ran the workspace's hook scripts. The DoltStorage decorator chain fires on_create/on_update/on_close after every mutation it lands; the unit-of-work plumbing — the one proxied-server mode writes through — fired nothing from the plumbing, so an integration wired to .beads/hooks/ silently missed every mutation that went through it. Four commands (update, close, reopen, gate) had grown hand-wired hook calls of their own to paper over the gap; every other write — create, batch create and close, claim, comment, label edits, dependency edits — ran no hook at all. A notifying wrapper now fires them from the plumbing, buffered during the transaction and drained only after the commit succeeds, so a rolled-back or retried write reports nothing.

    Six changes are visible in proxied mode, and all six move it toward what the embedded path has always done (a seventh, below, applies to both):

    • One firing site per plumbing. The four per-command hook calls are gone.
    • bd update -s closed fires on_update only. It used to fire on_close as well. bd close is what fires on_close, on both plumbings.
    • bd close no longer fires an on_update ahead of its on_close.
    • Hooks run fire-and-forget, after the write commits, rather than synchronously inside the command.
    • no-hooks is honored. The per-command calls ignored it, so hooks fired in proxied mode even when they were switched off.
    • An idempotent re-close fires on_close, and a double bd gate resolve fires it too. The per-command calls suppressed both; the embedded path has always fired for a close that succeeded, changed row or not, so a script reconciling on "it is closed" is told every time rather than only the first.

    And one change every workspace sees, embedded and proxied alike: a command now waits for its own fire-and-forget hooks before exiting, bounded by the per-hook timeout (10s). Both plumbings run hook scripts on background goroutines, and a short command could return from main before one had even started — a hook that silently never fired. The wait happens after the store closes, so a hook script's own bd can open the workspace, and it runs on the error exit paths too, so a partial batch still delivers the hook for what it did commit. A command that fires no hooks waits on nothing.

    bd serve still runs no hooks, and bd import still fires none on either plumbing (both import through the batch-upsert engine rather than the per-issue verbs).

  • --dolt-auto-commit batch/off now actually defer version commits in SQL-server mode (bd-4wamg). The mode was silently inert there: the CLI's auto-commit policy was embedded-only, and the storage layer minted one Dolt commit inside every write transaction regardless — measured at one commit per write through all five configuration paths (flag, config.yaml, env, bd config set, no-daemon). The write verbs now thread the batch/off deferral to the storage layer's commit sites, which leave writes in the working set for an explicit commit point (bd dolt commit), in server and embedded mode alike. The server-mode default changes spelling, not behavior: it used to resolve to off (while behaving like on); it now resolves to on, naming what a default server-mode write has always done. The --dolt-auto-commit help also stops claiming Default: off. On a shared server, staging is table-level, so an on-mode writer's commit may sweep up another client's deferred rows — deferral bounds who creates commits, not commit contents.

    Scope: embedded and direct SQL-server modes. Proxied-server routes never apply auto-commit policy (they short-circuit before it resolves), so batch/off remain inert there — tracked as a follow-up. A deployment that had explicitly configured dolt.auto-commit: off from the old help text note: in server mode that was behaving like on; it now genuinely stops minting version commits until an explicit bd dolt commit.

  • The no-ID "last touched issue" fallback on bd update / bd close is now interactive-only (bd-m00pb, #4839). Previously a scripted bd update $ID ... with an accidentally empty $ID silently mutated whatever issue was touched last — a real agent session corrupted an unrelated closed bead this way. The missing-ID refusal happens in argument validation, before any store open, migration, or auto-import side effect. The fallback now requires stdin to be a terminal; BD_NON_INTERACTIVE=1 and CI=1/true also disable it. Scripts that intentionally relied on it can set BD_LAST_TOUCHED_FALLBACK=1 (or =0 to disable the fallback everywhere). Explicit IDs and bd show --current are unaffected.

  • A dated defer now means what it says: the issue returns to the ready front when the date passes (bd-i8qx8). bd defer --until and bd update --defer set status=deferred plus a defer_until timestamp, but nothing ever flipped the status back — so an expired defer stayed invisible to bd ready forever until a human ran bd undefer. One deployment measured 241 issues (including P1s) silently dark this way, regenerating daily from correct-looking automation.

    Ready-front reads and claims (bd ready, bd ready --claim, bd list --ready, and the serve/proxied Reader roles) now run a lazy wake sweep first: every issue (and wisp) with status=deferred and defer_until <= now flips to status=open, defer_until=NULL — byte-identical to what bd undefer writes, so a later dateless re-defer cannot inherit a stale past date. Each wake records a status_changed event (actor bd-defer-wake).

    What does NOT change: a dateless defer (bd defer with no --until, defer_until NULL) is the indefinite icebox and never auto-wakes — bd undefer remains its only exit. The sweep is a no-op when nothing has expired (no dolt commit is minted), advisory by contract (a ready listing never fails because the sweep could not run — strict --readonly skips it), and identical in embedded, server, and proxied-server modes.

  • bd delete --cascade no longer marks LIVE issues as deleted in other issues' text (bd-x82so). When the deletion named a wisp, the set used to rewrite [deleted:<id>] citations was computed differently from the set that was actually deleted: dependents reachable only through a wisp root were left in place, and then their neighbours' description, notes, design and acceptance_criteria were rewritten to say those still-present rows had been deleted. bd wisp gc ran exactly this path, because it hardcodes cascade.

    Text already corrupted by earlier builds is NOT repaired. Those are stored bytes and nothing here scans for them; a [deleted:<id>] marker naming a row that still exists is a leftover from a pre-fix build. Searching for markers whose id is still present will find them.

    Three consequences of the same fix, all on the direct and embedded routes and all converging onto what the team-server route already did:

    • bd delete <wisp> --cascade now deletes the wisp's dependents in both planes, as --cascade has always claimed. It previously stopped at the wisp, reported only what it deleted, and gave no sign it had done less than asked. Anyone relying on that under-deletion will see more rows go — and bd wisp gc now removes durable rows hanging off abandoned wisps.
    • bd delete <wisp> with a durable dependent, unforced, is now refused with issue <id> has dependents not in deletion set. It previously succeeded and silently orphaned the dependent.
    • bd delete <wisp> --force now lists that dependent in the orphan report, where it previously reported nothing, and the removed cross-plane edge is counted.

    The two set computations are gone: one DeletionSet, resolved once, is handed to the neighbourhood read, the delete and the rewrite, so they cannot drift apart again.

  • bd query no longer drops matches from an OR or NOT query (bd-pohmh). Your results will change, and they change by getting bigger. To answer an expression the storage filter cannot express, both routes fetched max(3 × --limit, 100) rows and applied the expression to those in memory. A match past that window was absent from the page — and the truncation hint said nothing, so the answer looked complete. bd query "type=bug OR label=urgent" over a workspace with more than a hundred issues has been returning an arbitrary prefix of its answer.

    The window is gone: the expression is now evaluated against every candidate row, the page is the first --limit MATCHES, and the truncation hint is exact. The cost is real and worth knowing: a broad OR/NOT expression over a large workspace now reads every row its plain filters admit, where it used to read at most a few hundred. Narrow the expression — the parts of it that ARE plain filters still bound the read — if that matters to you.

    Two smaller changes ride along. bd query --offset N now works with OR/NOT expressions, where it used to be refused outright: the offset skips matches, which is only meaningful now that every match is seen. --offset with --sort is still refused, with new wording, and so is a malformed expression: both messages now come from one place instead of one per route.

  • bd purge --pattern and bd prune --pattern no longer report success on a malformed glob (bd-pn231). Both routes matched patterns with the error return of filepath.Match discarded, so bd prune --pattern '[' --force printed "No closed beads to prune" and exited 0 — indistinguishable from a correct pattern over an empty set. A pattern the glob syntax refuses is now an error naming the pattern, on both routes.

  • bd purge --dry-run and bd prune --dry-run no longer fail on sweeps the real run completes (bd-pn231). The preview asked for the refuse-if-any- external-dependent path while the confirmed run asked for the force path, so a dry run could report "has dependents not in deletion set" for a sweep --force finishes by orphaning them — and the CLI then swallowed the error and printed zeros. The preview and the run now ask the same question.

Added

  • GET /v0/beads/dependencies/blocking — the blocking decoration bd list prints, over HTTP (bd-5k21g). bd serve now answers ?ids=bd-abc&ids=bd-def with, per requested id, its open blockers, the beads it blocks and its open parent — the derived summary bd list renders as blocked by: and blocks:, from the same issueops.BlockingAnnotator both CLI routes call. capabilities gains dependencies.blocking.

    It is not a mode of GET /v0/beads/dependencies: that one returns stored edge rows and applies nothing, this one applies the status rule and separates the parent edge from the blocking ones. It is also not a page — the number of ids the caller named is what bounds it, so there is no limit and no cursor. An id that names nothing is annotated bare rather than reported missing, which is the same answer the CLI gives.

  • GET /v0/beads/dependencies/tree — the bd dep tree walk over HTTP (bd-lx5iu). bd serve now answers a recursive walk from one root — ?root_id=bd-abc&direction=both&max_depth=3&status=open — with the flat TreeNode list bd dep tree --json emits, from the same issueops.TreeWalker both CLI routes call. capabilities gains dependencies.tree.

    It is a sibling of GET /v0/beads/dependencies rather than a mode of it: that one answers raw stored edge rows for many anchors at one hop, this one recurses from a single anchor. A root_id that names nothing is a 404 here, where an unknown issue_id there is reported in missing — one anchor leaves no other answer to preserve. There is no max_rows: the defensive cap is a CLI circuit breaker and refusing a whole answer is not something this surface offers a remote client.

  • GET /v0/beads/issues:query — the bd query expression language over HTTP (bd-pohmh). bd serve now answers boolean expressions — ?q=type%3Dbug+OR+label%3Durgent — with the same page shape the ready listing returns, under the same completeness guarantee as the CLI, because both call one issueops.Querier. It is the only operation on the surface that takes a disjunction; every filter parameter of GET /v0/beads/issues narrows. There is no cursor and no offset: a predicate query's matching set is assembled outside the database, so there is no keyset position to encode. capabilities gains issues.query.

  • POST /v0/beads/issues:delete — deleting named beads over HTTP (bd-x82so). bd serve gains the second destructive operation on its surface: a body of ids plus cascade, force and dry_run, answering with the counts and the orphan list. It calls the same issueops.Deleter bd delete calls, so the dependents guard, the all-or-nothing id resolution and the reference rewrite are the library's rather than the handler's — the endpoint cannot orphan a graph by omission. cascade and force both default false, which is the guarded mode. capabilities gains issues.delete.

  • POST /v0/beads/issues:batchCreate — creating many issues over HTTP (bd-lu170). bd serve gains the second write on its surface and the first one that creates rows: a request carrying an actor and up to 100 items is created as ONE transaction, or none of it is. It runs on the new issueops.BatchCreator role that both bd create --file routes now call, and capabilities gains issues.batchCreate.

    The server assigns every id. There is no id member on an item, so this operation can never adopt or overwrite a stored row; the generated ids come back in items, in request order, which is the only place a client can learn them. bd import remains the upsert surface and is not published here.

    A dependency target may name an issue the workspace holds, an item created EARLIER in the same request, an external: reference, or an id belonging to another repository. Anything else is a 400 and nothing is created. Hooks do not fire and the per-command auto-commit machinery does not run, exactly as for POST /v0/beads/issues/{id}:claim.

  • POST /v0/beads/issues:sweep — bulk clearance of closed beads over HTTP (bd-pn231). The operation behind bd purge (tier: ephemeral) and bd prune (tier: durable), on the new issueops.Sweeper role both CLI routes now call. capabilities gains issues.sweep.

    This is the first DESTRUCTIVE operation on the surface, and nothing it deletes comes back. bd serve binds loopback, and its bearer — optional, and where configured a single shared token admitting a client to the whole surface — grants no narrower right, so anyone who can reach the port with a credential it accepts can now clear closed beads: --help gained a DESTRUCTIVE OPERATIONS section and the --allow-non-loopback warning says so. A durable sweep with neither a cutoff nor a pattern is refused by the role itself rather than by the handler, and dry_run: true costs the same request without writing anything.

    protect_referenced defaults to true here, where the CLI's equivalent (--ignore-references) is opt-out. The local flag and the remote member therefore disagree on purpose: being authenticated here says nothing about whether this particular deletion was meant — one shared token admits a client to everything published — so a remote caller who omits a member should not get a weaker guard than the operator typing the same sweep locally. Send protect_referenced: false to match the CLI default.

  • GET /v0/beads/config and GET /v0/beads/config/{key} — the workspace's stored settings over HTTP (bd-kzepq). bd serve now answers for the settings plane bd config list and bd config get read. Both operations run on the new issueops.WorkspaceConfig role the CLI reaches through the same accessor, and capabilities gains config.list and config.get.

    A setting whose KEY marks it as credential-bearing — the name contains token, secret, password or an API-key spelling, which notion.token does — is published with redacted: true and no value. Withheld values are OMITTED rather than masked, so a client cannot mistake a placeholder for configuration. The CLI is unchanged and still prints stored values in full: its caller already holds the database, while redaction is the whole control on this surface — a bearer here is optional and, where configured, shared and surface-wide, so it cannot withhold one value from one caller.

    There is no HTTP WRITE, deliberately. bd config set routes most keys to config.yaml or to git config before the database is ever reached, and both live on the client's filesystem, so a server-side write of (say) export.auto would return success and land a row nothing reads. The multi-source views — bd config show, drift, apply, validate — are off the role and off the wire for the same reason.

  • bd serve runs on a registered storage backend. A downstream distribution that registers a backend (github.com/steveyegge/beads/backend.Register) could already use it from every ordinary bd command — bd list, bd create and friends dispatch the store open through the registry — but bd serve alone reimplemented a creation path of its own, hard-wired to Dolt's SQL wire, and refused the workspace before it got there. It now uses the store the root command already opened, through the same registry dispatch: zero creation calls of its own, so there is one creation path rather than two. The two issue roles are taken from BENEATH the workspace's hook decorator, keeping the documented "hooks do not fire" contract (httpapi.Listen refuses a hook-firing role rather than trusting the caller), and the store's lifecycle stays where it was — opened and closed by the root command, closed only after the server has drained. The startup line names the source (db=roles) and the backend (mode="<name> (registered backend)").

    The permanent refusal is unchanged and unweakened: an embedded-Dolt workspace is still refused, because its commit protocol runs outside the SQL transaction and this server's per-request atomicity would be a lie there. The classification consults the registry before any Dolt-mode signal, which is the order the store open already resolves them in — and which also fixes a latent CGO_ENABLED=0 bug where a registered workspace was handed to the Dolt provider and either failed with a misleading Dolt error or connected to a defaulted host and served the wrong database.

Changed

  • BREAKING: bd config list and GET /v0/beads/config no longer enumerate the kv. plane, which is where bd remember memories live. Memories and generic bd kv values are stored as config rows under kv. — user data riding in the settings table because there is one table, not because they are settings — and both doors onto the settings plane were listing them with their values. On the HTTP door that meant an unauthenticated GET /v0/beads/config handed every stored memory to anything that could reach the port, and the surface's redaction is no defence there: it decides on the KEY name, while a memory's content is in the VALUE, so a memory containing a credential under an innocuous slug was served verbatim.

    The exclusion is in the shared role both doors call, not in the HTTP handler, so the CLI and the API cannot drift apart on what a setting is. The whole kv. prefix goes, not just kv.memory..

    What still works: bd memories, bd recall and bd kv list are unaffected — they read the store directly and are the purpose-built views over those rows. So are bd prime's memory injection and export's memory extraction. Point access is unchanged and deliberately so: bd config get kv.foo and GET /v0/beads/config/kv.foo still answer a key named exactly, and bd config set / bd config unset still write and remove one. The enumeration was what made a memory discoverable; naming one you already know is a different question, and refusing it would take away the escape hatch of removing a wedged memory through bd config unset.

    What breaks: a script parsing bd config list or its --json output for kv. rows. Generic kv. rows have no other HTTP view today.

  • BREAKING: bd --readonly serve is now refused instead of binding a server that cannot do what it advertises. On a Dolt SQL-server workspace this previously bound a fully functional, fully writable server — the flag was a silent no-op there — so anyone scripting bd --readonly serve as a "safe" read server loses it on upgrade and should drop the flag. The advertised capability set is a property of the build and includes the issue-claim operation, so under strict readonly the two database sources degraded differently and silently: a registered backend bound on its read-only store and answered every claim with an opaque 500 while still advertising issues.claim, and a Dolt SQL-server workspace built its own writable provider and let every claim land, so the flag bought nothing. bd serve now refuses to start under --readonly (or readonly in config), before it resolves the workspace, so the answer is the same on both.

Fixed

  • GET /v0/beads/context and bd context no longer describe every workspace as Dolt. The shared identity projection hardcoded backend: dolt and copied dolt_mode and database unconditionally — and both of those default rather than fail, so a workspace on a registered backend was reported as backend="dolt", dolt_mode="embedded", database="beads": a confident description of a topology it is not on, on the one endpoint automation is told to trust for a server's identity. The backend is now named as the store open resolves it, and the two Dolt-derived values are reported only for the Dolt backend; a registered backend reports the empty string for both, which is the only value bd can assert about a backend it does not implement. No wire field is renamed, retyped or dropped — both remain required strings, and the v0 shape is unchanged.

  • Public surface for out-of-tree storage backends (bd-h3dib.2). The storage backend contract and its conformance suite are now importable by external Go modules — the conformance-gated external-backend path promised on #4415. The new github.com/steveyegge/beads/backend package exports, by type alias, the engine interface (backend.DoltStorage) and every type its 164 method signatures reach, the sentinel errors the contract requires by identity, and thin wrappers over the backend registry (backend.Register); the suite moved from internal/storage/conformance to github.com/steveyegge/beads/backend/conformance unchanged, so an external backend proves itself with the same conformance.RunAll the embedded-Dolt reference runs in CI. Aliases mean identical types: a store compiled against the public names satisfies the internal interface, and errors.Is matches across the module boundary. A reflection-based drift guard (backend.TestPublicSurfaceComplete) fails in-tree if a contract change ever reaches an internal type with no public alias. Stability: EXPERIMENTAL — the contract still evolves; pin an exact version and re-run the conformance suite on every bump.

  • bd serve — the beads work surface over HTTP (bd-serve v0). Automation clients and orchestrators that fork a bd subprocess per call can now hold one connection instead. bd serve binds loopback and answers six OpenAPI-specified operations: GET /healthz, GET /v0/beads/context, GET /v0/beads/ready, GET /v0/beads/issues, GET /v0/beads/issues/{id}, and the one write, POST /v0/beads/issues/{id}:claim. The wire contract is internal/httpapi/spec/openapi.v0.yaml; types are generated from it and make api-check fails a change that edits one without the other. GET /v0/beads/context reports which operations the running build actually implements, derived from the registered handlers, so a release cut mid-slice cannot advertise one that does not work.

    The reads run on issueops.Reader, the same role bd show --json reaches, so filter construction, workspace config, default limits and the wisp fallback happen inside the role rather than being re-performed per front door. The exact scope of that property — what is shared, what is machine-enforced and by which lint rule, and what is not enforced — is stated in full in issueops.Reader's doc comment and in engdocs/design/bd-serve-v0.md; it is narrower than "the CLI and the API cannot disagree", and the difference is written down.

    Errors are typed. Every non-2xx is an RFC 9457 problem+json document with a machine-readable code (invalid_argument, invalid_cursor, not_found, already_claimed, not_claimable, busy, db_unavailable, internal), one frozen HTTP status per code, and Retry-After on the retryable ones. A client that classified claim conflicts by substring-matching error text should switch to the typed 409: assignee and issue_status come from a read inside the losing transaction, never from parsing the sentinel's prose. A 5xx detail is a fixed string per code — driver errors embed the DSN — and the body's request_id correlates to the one log line that carries the real error.

    Paging is an opaque keyset cursor on GET /v0/beads/issues (GET /v0/beads/ready has none: its sort policies admit no keyset predicate). The token holds a position and a private encoding version and nothing else, so it does not expire, survives a restart, and has exactly one failure recovery — restart paging without it. It also carries no filters, so a cursor reused under changed filters is not refused and silently resumes from the old position; repeat every filter verbatim for a traversal.

    Deliberate omissions, stated as contract rather than gaps. No authentication and no TLS — the trust model is the loopback boundary the database already relies on, and --allow-non-loopback is an operator decision that additionally refuses unlimited (limit=0) reads. A DNS-rebinding Host allowlist with no off switch. Hooks do not fire on an HTTP claim (a user-controlled subprocess per mutation is an unbounded latency multiplier and an orphaned child at shutdown). The per-command auto-commit machinery does not run: durability is per request, and an idempotent re-claim by the current holder writes no commit at all, so a polling client cannot mint empty storage commits. An HTTP actor is caller-asserted provenance for the audit trail, not authenticated identity — the same thing --actor has always been.

    Embedded-Dolt workspaces are refused permanently: that backend commits outside the SQL transaction, so the per-request atomicity this contract states could not hold there.

    Operators: pass an explicit port. The default --addr 127.0.0.1:0 takes an ephemeral one and carries no mutual exclusion — N servers run side by side on different ports. Probe readiness with GET /v0/beads/ready?limit=1; /healthz is liveness only and stays green while the database is unreachable. Deployment, the connection budget for a shared dolt sql-server, the request-log format and the ambiguous-shutdown re-claim recovery are in engdocs/SERVE_RUNBOOK.md.

  • Replica-aware leases: bd reclaim no longer reverts a lease another replica granted (wy-jpd3.7). A lease is only meaningful on the replica that granted it — the other machine's liveness view is stale by up to one sync interval — so each lease now records its granting node (leases.granted_node, ignored migration 0016; types.Issue.LeaseGrantedNode, which rides the JSONL interchange so an imported lease keeps its provenance). bd reclaim skips a lease that positively names a different replica, naming each skip on stderr, and the new bd reclaim --any-replica (types.ReclaimFilter.AnyReplica) is the escape hatch for a replica that is permanently gone. bd show annotates a lease granted elsewhere.

    The guard is opt-in and fail-open. It arms only where you set the new node_id config key (BEADS_NODE_ID / BD_NODE_ID, or bd config set node_id <name>, which writes the per-machine ~/.config/bd/config.yaml — a node_id in the git-tracked project .beads/config.yaml would give every clone the same identity and leave the guard armed but inert). There is deliberately no hostname fallback: the hostname names the client process's machine, not the store, so on a shared or remote dolt sql-server — where many hosts are clients of ONE store and there is no sync interval to defend against — it would stop a supervisor reaping any worker's lease at all. Unset (the default), and a lease of unknown provenance, behave exactly as before, so an upgrade can never strand a lease the reaper could previously recover. Also documents the invariant the guard cannot enforce: grace window > sync interval, and lease TTL > sync interval.

  • bd sync — one verb for the federation loop (wy-jpd3.4). Every multi-machine beads deployment hand-rolls the same sequence in shell; this ships it. bd sync [--remote <name>] [--attempts N] pulls, checks for merge conflicts, runs the full is_blocked recompute, and pushes — retrying a bounded number of times (default 3) when another replica wins the push race. Two properties are the point of the verb. Conflicts are detected positively, from the merge's own captured conflict rows and from dolt_conflicts, never inferred from the pull's exit status: a pull fails for plenty of reasons that are not conflicts, and a settled merge that is conflicted aborts and leaves dolt_conflicts empty, so an exit-status guess invents phantom conflicts and misses real ones. And a conflict sync cannot settle is never resolved by picking a side: it halts before recomputing or pushing and exits 2, with no --strategy-style override to make it do otherwise. (The pull underneath still auto-settles the convergent classes it always has — machine-local metadata, audit-only dependency rows, LWW on issue cells; anything past those halts.) The halt message reports which state the halt left behind, read from which detection source fired rather than assumed: the SQL pull route aborts the merge and restores the working set, while the CLI/git-protocol route deliberately leaves the conflict rows live. The recompute between pull and push is not bookkeeping either: is_blocked is denormalized, so a merge that brings in a dependency edge from another replica leaves bd ready stale until it runs. It runs unconditionally, on every attempt — RecomputeAllBlocked is precisely the repair that does not depend on a merge advancing HEAD, so gating it on "did anything merge" would mean a column left stale by a hand-resolved conflict (a state bd sync creates by exiting 2) is never repaired again while every tick reports success. Exit codes are the machine contract, so a sync timer can branch without parsing output: 0 synced, 1 error, 2 merge conflict (halted, nothing pushed), 3 push-race retries exhausted (transient — retry next tick). --json reports the outcome as {"status", "attempts", "conflicts", "conflicts_live", "rows_corrected", "pushed"} on every non-error exit; exit 1 emits bd's standard {"error": ...} envelope. A confirmed-no-remote rig exits 0 with the same guidance bd dolt push prints; dolt.local-only and no-push are honored.

  • bd reclaim scope filters (wy-jpd3.3). bd reclaim gained --label / --label-any / --exclude-label / --assignee / --id, mirroring the claim-side label surface (bd ready --claim), so a supervisor can scope its reaper to exactly the partition it claims from. Filters AND-combine and never widen the stale set. This matters on a federated deployment: each replica's view of another machine's liveness is stale by up to one sync interval, so an unscoped reaper on one machine can revert a unit that is alive on the machine that granted its lease — point each supervisor's reclaim at its own claim partition and that cannot happen. A scope flag supplied with no usable value (e.g. --label "$LANE" with LANE unset) is a hard error rather than a silent degrade into a global sweep, and the builder fails closed (1=0) for programmatic callers. bd reclaim --json now reports a scoped boolean so a reclaim log distinguishes a scoped sweep from a global one.

  • Every command that had a role now reaches it through the accessor, on both routes (facade waves 1-4). This is the close of the programme: twenty-three capability accessors on the storage substrate, each answering a role interface in the leaf package issueops, each with a backend-independent conformance suite run against the Dolt server store, the embedded Dolt store and the unit-of-work provider, and each reached by the CLI's direct route, the CLI's --proxied-server route and — where the capability is published — one HTTP operation, all calling the same role. bd serve publishes sixteen capability-bearing operations, twelve of them added by the programme, every one written into internal/httpapi/spec/openapi.v0.yaml before its handler existed.

    What did NOT get a role is as deliberate as what did. bd show and bd close were already behind Reader and BatchCloser/Lifecycle on both routes, so there was nothing to converge. bd version prints build information and a role is an interface onto a substrate — the one part of it that touches state, the recorded-version marker, got VersionReconciler instead. Most of bd init CREATES the substrate a role would be reached through; only the workspace identity is reachable as a store capability, and that is what Bootstrapper and InitVerifier are. bd create --graph is deferred on the record: it is an UPSERT with caller-supplied symbolic names and BatchCreator is create-only with server-assigned ids, so folding them would have meant one role with two incompatible id policies.

  • A shared semantic contract for every issue role (bd-yby99). Seven of the eight role interfaces in issueops now have a backend-independent conformance suite in backend/conformance, run against the Dolt server store, the embedded Dolt store, and the unit-of-work provider. One hundred and twenty-nine cases cover Reader, Relations, Commenter, DependencyEditor, ReadyClaimer, BatchCloser and Lifecycle — the last spanning Create, Update, Close and Reopen. The unit-of-work leg runs 127 of them: the two staging cases assert which concrete Dolt tables a commit staged, which that provider has no way to observe. Each case is written against the promise stated in the role's doc comment, cited by file and line, rather than against any implementation's current behaviour, so the doc comment is the artefact under test as much as the code is.

    Read the three legs for what they are. The Dolt and embedded-Dolt stores share one validate/execute body for six of the roles and the entire implementation for Reader, so a green run on all three is two readings of the contract plus a wrapper-and-engine check, not three independent votes. The unit-of-work provider is the only separately written body, and every divergence found so far where one backend disagreed with another rather than all three with the doc has surfaced there.

    Where an implementation contradicts its own doc comment the case still asserts what the doc says, and the disagreeing backend is parked at its wiring site with a greppable KNOWN DIVERGENCE <bead> skip, so the agreeing backends stay pinned. Thirty-one findings have come out of that, all children of bd-yby99: seven divergences, where the implementations disagreed, and twenty-four spec gaps, where the doc was silent rather than wrong and the case carried an in-file SPEC-GAP <bead> marker instead of an invented assertion.

    Every parked case now runs. The three findings that had a backend parked are adjudicated and fixed, so the five KNOWN DIVERGENCE skips are gone and all three legs run their full set with zero skips: an invalid ClaimNext sort policy now matches ErrValidation (bd-yby99.18), a reparent on the unit-of-work provider now replaces every parent edge (bd-yby99.26), and a dependency removal on that provider now reaches the ephemeral plane (bd-yby99.17). Each has its own entry below; none was resolved by weakening a case.

    Twenty-two of the spec gaps are adjudicated as well. In each the leaf doc gained the promise it had been missing and the contract case was tightened to assert it, in that order, because the doc is the spec and a case may never assert more than the doc promises. Ten of the adjudications moved production code rather than only the doc — bd-ocrn7, .3, .4, .7, .8, .17, .18, .19, .20 and .26 — and each is a user-visible change with its own entry below, not a quiet edit made to turn a case green.

    Three findings remain open, and none is parked — no case skips anywhere. bd-yby99.9 is the doc's own admitted silence, named in the leaf text so a caller does not read the silence as a promise: the dependency-existence refusals still carry no error identity. bd-yby99.1 is with bd-soezd for measurement rather than adjudication. bd-yby99.30 records a CLI retry-safety property the bd-yby99.19 decision traded away (see below). The empty-string label question (bd-yby99.28, bd-yby99.29) is closed, not open: the entry is now dropped on every backend and LabelPatch says so. Five duplicated per-backend test files were deleted outright and a sixth trimmed, their scenarios folded into the shared cases; the Dolt-only facts they also pinned — history-entry message spellings, working-set cleanliness, the GH#2455 no-op removal sweep — stay Dolt-only in persistence tests.

Changed

  • bd delete --force on a TEAM SERVER now orphans dependents instead of deleting them (bd-x82so). Read this one before you upgrade a team-server script.

    Against a proxied server, bd delete X --force used to delete X and its whole dependent subtree, and --cascade was refused outright with "not supported in proxied-server mode (delete always cascades)". There was no way to delete one bead on a team server without taking everything that depended on it, and asking for the safer behaviour was an error.

    Both routes now mean the same three things by the same two flags, which are the local database's meanings, unchanged:

    you type you get
    bd delete X REFUSED if anything outside the request depends on X
    bd delete X --force X is deleted; its dependents survive, ORPHANED
    bd delete X --cascade --force X and its whole dependent subtree are deleted

    A team-server script that relied on the implicit cascade will start leaving orphans. Add --cascade to it. The proxied --json result gains orphaned_issues so a script can see them; a cascade leaves it null.

  • Proxied bd init ADOPTS an existing project identity instead of overwriting it (bd-zl3u8). This one silently corrupted shared databases. Against a team server whose database another rig had already identified, bd init --proxied-server wrote its own locally-derived issue prefix and project id straight over the stored pair — renaming every id the co-tenant was about to mint. It now reads the identity first and prints "Adopted project identity from existing database", which is what the direct route has always done. There is no flag that restores the overwrite.

    Two consequences ride along. bd init no longer fills in a missing project id on a substrate that already carries an issue prefix: half-identified is now adopted as it stands rather than completed, because the guard that stops the overwrite above cannot tell a half-provisioned workspace from a shared one mid-write. Re-identifying is an explicit act with no command yet; if you have such a workspace, it keeps working and it stays half-identified. And bd init with BEADS_DIR pointing outside the repository now records a project id, taken from that workspace's own metadata.json, where it previously recorded none and left exactly the state the new refusal will not complete.

  • bd list and bd children print blocker ids in ASCENDING ORDER (bd-5k21g). The blocked by: and blocks: ids in compact and agent output came out in query order, which is map-iteration and query-plan dependent, so an unchanged workspace could print different bytes on two consecutive runs. They are now sorted ascending with repeats collapsed, on both routes. Only rows with two or more live blockers change; a diff-based script that pinned the old order was pinning an order the database never promised.

  • bd delete <one-id> without --force now REFUSES when the bead has an outside dependent, instead of printing a preview and exiting 0 (bd-x82so). This is the direct route converging with itself: the multi-id path and --dry-run have always refused here, and only the unconfirmed one-id preview did not. The refusal names the bead and says to use --cascade or --force, which is the same sentence the other two paths already printed.

  • bd delete rewrites text references INSIDE the delete transaction (bd-x82so). The direct route deleted the rows in one transaction and then rewrote [deleted:ID] into the neighbouring beads' descriptions afterwards, discarding every update error it got; a failure in the window left a workspace whose beads were gone and whose descriptions still cited them by id. The whole operation — the existence probe, the dependents guard, the cascade expansion, the deletion and the rewrite — is now one transaction on the new issueops.Deleter role, shared by both routes and by bd serve.

    The trade, said out loud: the transaction is now as large as the deletion plus its graph neighbourhood, so a delete whose neighbourhood is big enough to exceed the backend's write timeout fails whole instead of deleting the beads and leaving the text stale. Split a very large --from-file batch.

  • bd delete refuses the whole request when any id names no bead (bd-x82so), on both routes and under --dry-run. The multi-id path already did; the guarantee is now the role's, so nothing partial is deleted before the typo is reported.

  • bd dep tree --direction=both now sees ONE database state (bd-lx5iu). Both routes made two independent reads — the dependency walk and the dependent walk — with no transaction spanning them, so an edge added between them could leave the two halves of the picture describing different graphs. The whole walk, both directions, now runs inside one transaction on the new issueops.TreeWalker role.

  • bd dep tree on a team server resolves partial ids and honors --max-rows (bd-lx5iu). Two things the --proxied-server route did not do and the direct route did.

    bd dep tree a1b2 used to fail there: the argument went to storage verbatim, so the abbreviation that worked on a direct workspace did not work on a shared one. It resolves now, and an id that resolves to nothing reports the same lookup failure the direct route reports.

    bd dep tree X --max-rows N used to be refused outright — "not supported in proxied-server mode" — because the unit-of-work path threaded no cap. The cap now rides the role's request, so all three backends enforce it and the flag means the same thing wherever the command runs. It is still POST-HOC: a tree walk has no query filter to thread a cap through, so the full tree is built and then counted.

  • bd dep tree --show-all-paths is documented as the no-op it always was (bd-lx5iu). Nothing has ever read the flag. It was accepted and passed down to a walk that ignored it, so a node reachable by two paths has always been shown once, under the first path that reached it. The flag is still accepted so no script breaks, its help text now says it is ignored, and the one-visit rule is a stated promise of the role rather than an accident of the walk. TreeNode's truncated member is dead in the same way: it is always false, because a node beyond --max-depth is absent from the answer rather than present and flagged.

  • bd purge and bd prune select and delete in ONE transaction (bd-pn231). The direct route used to search in one transaction and delete in another, so a bead closed between the two could be counted and not deleted, or deleted and not counted. Both routes now run the whole sweep as one act on the new issueops.Sweeper role, and the count a sweep reports is the set it deleted.

    The trade is worth knowing before you run a very large purge on a team server. One transaction means the Dolt server backend can no longer batch wisp deletions 200 at a time, so a purge big enough to exceed Dolt's write timeout now fails whole instead of deleting part of the workspace and reporting nothing. Sweep in slices (--pattern, --closed-before) if you are clearing a very large backlog.

    The version-control entry for a sweep is now bd: sweep <n> <tier> bead(s) on both routes, where the proxied route wrote bd: prune <n> bead(s) and the direct route wrote bd: delete <n> issue(s), naming neither the command nor the tier. Scripts that grep commit messages for the old spellings need updating.

  • bd list --max-rows reports its own malformed value first (bd-b34o7). The row cap now rides the list request rather than being stamped onto the filter after it was built, which moves the flag's validation ahead of the filter's. bd list --max-rows -1 --status bogus now reports the --max-rows usage error where it used to report the invalid status; the exit code is unchanged and it takes two bad arguments to reach.

  • bd create --file is all or nothing, and its plan-wide flags now mean something (bd-lu170). Both routes build one issueops.CreateBatchRequest and ask for the new BatchCreator role, so a markdown plan is created as one act with one history entry on either.

    A file the workspace refuses now creates NOTHING. Each template goes through the same validation a single bd create goes through — the workspace's configured types and statuses, the field lengths, the create-only guard — and a template that fails takes the whole file down with it. Re-running a file that used to fail can therefore give a different outcome than it did before, because the run before it may have landed rows this one refuses to duplicate; check bd list before re-running a plan that errored on an older version.

    A declared dependency on an id that does not exist is now an error rather than a silently dropped edge. The direct route used to create the issues and leave the edge out, with nothing on stdout or stderr saying so, so a plan file with a typo produced a graph that looked complete and was not. An external: reference and an id belonging to another repository are still written, exactly as bd dep add accepts them; only a target that shares the workspace's own prefix and names no row is refused.

    --ephemeral, --no-history, --mol-type and --validate are honoured on the direct route. All four were accepted and silently ignored there while the proxied route acted on them, so bd create --file plan.md --ephemeral created durable issues on a local workspace and wisps on a team server. --validate (and validation.on-create) now lints every template on both routes, which is the policy a single-issue create has always applied. Issues created from a file also carry created_by and owner, as a single-issue create does.

    bd create --file --json emits the STORED rows on both routes. Each entry is the post-create snapshot the role reads back — with owner, created_by and dependency records carrying their own created_at, created_by and metadata — where the direct route used to marshal the structs it had built in memory. Additive: no member was removed, and both routes now emit one shape.

    Proxied --file loses its config.yaml fallback for types.custom, the same convergence onto the direct route's database-only vocabulary that landed for single-issue create (bd-7oyh5).

    bd create --graph is untouched: an upsert over a declared shape is a different contract, and it gets its own role when it gets one.

  • Proxied bd ready --json --limit N publishes pagination.total (bd-s10oa). Under BD_JSON_ENVELOPE=1 the direct route has always carried the size of the ready front beside a truncated page; the proxied route emitted no total key at all, so the same script read a number against a local workspace and nothing against a team server. Both routes now answer from the new issueops.ReadyCounter role. The count is only asked for when the page came back full, so an untruncated listing costs nothing extra.

  • Proxied bd dep list a b c reports ids that name nothing (bd-nhsno). The route handed its raw arguments to the batch read, whose answer simply has no entry for an id nothing matched, so a typo printed <id> has no dependencies — a clean graph — instead of a warning. It now prints warning: no issue found: <id> (skipped) on stderr, the line the direct route has always printed, and omits the anchor from stdout. Emitted in --json mode too, on stderr, so the document stays a flat array.

    Two smaller shape changes ride along, on both routes: a repeated anchor is answered ONCE, at its first mention, rather than twice; and bd dep list --type X a b prints <id> has no dependencies instead of a section header with no rows under it for an anchor whose every edge the filter rejected.

  • bd status --assigned reports why it failed (bd-4ze25). The direct route swallowed every underlying error and printed the causeless failed to get assigned statistics. Both routes now report the same underlying error. Relatedly, the proxied --no-blocked warning is derived from the ANSWER rather than printed unconditionally before the query, so it is correct for any backend and disappears by itself when a backend gains a fast path.

  • bd dep cycles output is deterministic, and no longer shortens a cycle it cannot fully describe (bd-wfkbv). Two changes, both visible in bytes.

    The report is now CANONICAL: each cycle's members are rotated so the lowest id comes first, and the cycles are sorted against each other. Two runs against an unchanged database previously disagreed — the walk iterated a Go map, so both the order of the cycles and each one's starting point moved between runs, --json included, and on a graph with overlapping cycles even the SET of cycles could differ. Diffing two snapshots is now meaningful.

    A cycle whose member has no row behind it is now reported HONESTLY. Members that could not be looked up were silently dropped from the path, so a three-node cycle rendered as a two-node one and looked complete, and a cycle none of whose members could be looked up vanished from the report and from the Found N dependency cycles count. Every member is now carried with its id, its description is omitted rather than the member, and the cycle is marked partial. The count no longer shrinks because rows went missing.

    --json therefore emits a NEW SHAPE: an array of {"members": [{"id", "issue"?}], "partial": bool} where it used to emit an array of arrays of issues. The issue member is the same issue object as before and is absent — never null — for a member this workspace holds no record of. Scripts that read bd dep cycles --json need updating.

    The post-add cycle warning that bd dep add, bd dep --blocks and bd link print is the same sweep and gains the same two properties, on both the direct and the proxied route.

  • The v0 HTTP surface reaches twelve operations (facade wave 2). This release adds six: GET /v0/beads/stats (bd-4ze25), GET /v0/beads/dependencies/cycles (bd-wfkbv), GET /v0/beads/dependencies (bd-nhsno), GET /v0/beads/ready:count (bd-s10oa) and the two config reads (bd-kzepq). Every one answers from the SAME issueops role its bd command reaches, so the two surfaces cannot drift: the operation is written into internal/httpapi/spec/openapi.v0.yaml first and the wire types are generated from it. capabilities now carries ready.list, ready.count, issues.list, issues.get, issues.claim, stats.get, config.list, config.get, dependencies.cycles, dependencies.list.

  • GET /v0/beads/stats — the workspace summary over HTTP (bd-4ze25). The same envelope bd status --json prints, minus recent_activity, which no shipped code path populates. skip_blocked is a HINT: a backend with no cheaper path answers with the full numbers and says so through blocked_count_skipped, which is derived from the answer rather than echoed from the request. assignee scopes the summary to one actor and is refused when empty rather than answering with the unassigned rows.

  • GET /v0/beads/dependencies — stored edge rows for several issues at once (bd-nhsno). The shape bd dep list a b c --json emits. An id that names nothing is reported in the response's missing member rather than turning the whole call into a 404, so a batch keeps the answers for the ids that WERE found; an edge whose target this database holds no row for is still an edge and is returned as stored.

  • GET /v0/beads/ready:count — the size of the ready front (bd-s10oa). Takes the same filter vocabulary as GET /v0/beads/ready, decoded once by the same code so the two cannot admit different parameters, and takes no limit and no sort: a cardinality has no page and no order.

  • bd serve publishes GET /v0/beads/dependencies/cycles (bd-wfkbv). The cycle sweep is now an HTTP operation as well as a command, answering from the same role, and capabilities gains dependencies.cycles. It takes no parameters and is not paginated: truncating would shrink the count, which is the number an operator acts on. bd link on a team server also runs its cycle sweep and its title lookups AFTER the write commits rather than inside it, so the warning describes a graph other clients can see.

  • bd count is one implementation again (bd-m8hn7). The command had two: the direct route opened its own store and the proxied route re-parsed the same flags against a unit of work, and each independently loaded the workspace list configuration before applying --include-infra. The cardinality contract bd count keeps with bd list was therefore asserted once and implemented twice. Both routes now go through the new issueops.Counter role, which builds its filter with the same shared builder bd list uses. Text and --json output are unchanged on both routes — the committed protocol corpus fixture for bd count is byte-identical — so this is a change you should only notice if the two routes previously disagreed. Two behaviours are now written down as promises rather than left as accidents: an unrecognised --status or --type still matches nothing and answers 0 rather than refusing (where bd list validates both), and a grouped count's Total is computed independently of its buckets, so overlapping dimensions such as labels do not sum to it.

  • Proxied bd create, bd reopen and bd update emit the same --json shape as their direct counterparts (bd-7oyh5, bd-1f5a2, bd-xt6de). Each of the three now answers from the hydrated post-state snapshot the issueops.Lifecycle role returns instead of a bare row read, so all three gained the labels field they were silently dropping, and — as on the direct route — dependency records are not included. Scripts that read proxied output should expect the direct route's documented shape; nothing was removed. bd update --claim --json is the one exception and still strips labels.

  • Proxied multi-id bd reopen records one history entry per id (bd-1f5a2). It wrote a single combined entry covering every id in the batch; the direct route has always written one per id, and both routes now do. On a team server a bd reopen a b c is consequently three round trips and three Dolt commits rather than one, which a script reopening a large batch will feel.

  • Proxied bd create validates the way the direct route validates (bd-7oyh5). Its custom issue-type check no longer falls back to config.yaml when the database has no custom types, its explicit-id prefix guard no longer consults the YAML prefix overlay, and its --type, --status and prefix refusals are worded as the role words them. The direct route has never had either fallback, so this is convergence onto the stricter of the two. The version-control entry a proxied create records also changes text, from bd: create <id> to create issue: the role publishes no field for a caller-supplied label, and no test asserted either string.

  • Proxied bd update error messages no longer name the stage that failed (bd-xt6de). "opening unit of work" and "committing" collapse into the generic per-id update failure, and an update whose commit fails with a context cancellation now aborts the batch instead of recording a per-id verdict for it. This matches the choice bd update --claim already made when it moved to the role. Exit codes are unchanged.

  • An empty-string label is dropped rather than stored (bd-yby99.29). Adding or replacing labels with an empty value wrote a labels row keyed on '' on the Dolt-backed stores and dropped it on the unit-of-work one, so the same edit left different rows depending on which route served it. Dropping wins on both routes now: an edit carrying only an empty value changes nothing and reports no change, and an empty value alongside real ones drops only itself rather than failing the edit. A row that renders as nothing and matches nothing was never worth storing, and refusing the whole update would have let one stray value sink an otherwise-good multi-label edit. An empty row already stored is left alone by add and remove, and cleared by a replace that omits it — ordinary set behaviour, not a migration, so nothing sweeps existing data.

  • An invalid sort policy is now a typed validation refusal (bd-yby99.18). Asking for a sort the ready query does not have — --sort newest, or a ClaimNext sort on a batch close — was refused with a bare error carrying no sentinel, on all three backends. A caller doing errors.Is(err, issueops.ErrValidation) to separate "you asked for something impossible" from "the database is unwell" classified it as a backend failure, which is the wrong half of the role's own error trichotomy. It now wraps ErrValidation. The message text is byte-identical — the wrap uses %.0w deliberately — so the HTTP surface that prefix-matches it to render a 400 rather than a 500, and every script reading the printed line, are unaffected.

  • A batch close that changed nothing no longer claims work or records history (bd-yby99.19). BatchCloser couples ClaimNext to "at least one item closed", and "closed" was being read as "the item returned no error" — which an already-closed item does. A batch whose items were all already closed therefore mutated nothing and still handed out a claim and wrote a history entry. Landing now means changed: an all-idempotent batch earns no claim and records no entry. A batch that closed even one item is unaffected.

    This has a cost worth stating plainly, tracked as bd-yby99.30. An agent that closed an issue, crashed before receiving its next work item, and retried bd close <id> --claim-next used to get work from the retry; it now gets nothing, because the retry closes nothing. That retry-safety property was deliberate, and the way to get it back without re-coupling a claim to a no-op is for cmd/bd to call ReadyClaimer.ClaimNext itself when the batch reports no change. That is a design call above the role seam and has not been made.

  • A batch's Dolt commit message no longer names wisp ids (bd-yby99.20). A batch may close ephemeral and durable items together, and the durable commit message named every item that landed. Wisp tables are dolt-ignored, so that message shipped a reference to work no other machine can resolve — the same federation leak Commenter already guards against. The message is now composed from durable landings alone and reports ephemeral ones as a count. Closing a wisp still counts as work landing: a wisp-only batch still succeeds and still earns its claim.

  • Breaking: bd list --ready now refuses a filter it cannot honor instead of silently ignoring it (bd-yby99.8). The --ready arm reaches the blocker-aware query through a narrower filter than the rest of bd list uses, and everything that filter could not carry was dropped without a word: bd list --ready --id tst-d68 answered with every ready issue in the workspace, not the one named. It now fails with a validation error naming the fields it could not honor.

    The refused combinations are --ready together with any of --id, --title, --spec, --title-contains, --desc-contains, --notes-contains, --external-contains, --external-ref, any --created/--updated/--closed/--defer/--due bound, --deferred, --overdue, --empty-description, --no-labels, --no-parent, --pinned, --priority-min, --priority-max, or an HTTP list cursor. A script that passed one of these was already getting the wrong answer; it now gets an error instead of silently wrong rows. Everything --ready does carry is unchanged — --type, all the label flags, --assignee, --no-assignee, --priority, --parent, --mol-type, --wisp-type, --metadata-field, --has-metadata-key, --exclude-type, --include-gates, --include-infra, --limit, --offset, --max-rows, --sort and --reverse — as is --no-pinned, which the ready query already satisfies by construction.

    What --ready carries, what it refuses, and the two things the ready set decides for itself (it never returns pinned issues, and it applies no template predicate of its own) are now stated on issueops.ListRequest.ReadyFlag and pinned by the Reader conformance contract at all three backends.

  • issueops.Reader now refuses an Offset it cannot honor instead of silently ignoring it (bd-yby99.7). The store-backed reader — one body, handed back by both the Dolt server store and the embedded store — renders LIMIT without OFFSET, so a caller that paged with Offset got the unpaged first page again and again with no error to notice: three matching rows came back 3/3/3 at Offset 0/1/2. Ready and List now return a typed *issueops.ErrUnsupported naming the operation and the backend. The unit-of-work reader, which does render OFFSET, keeps honoring it, and both behaviours are now pinned by one conformance case: an Offset is honored or refused, never ignored.

    No CLI behaviour changes. bd ready --offset, bd list --offset and bd query --offset already rejected a non-zero offset on the direct route and already reached the unit-of-work path under --proxied-server, where offset paging works and is unaffected. This closes the same hole one layer down, for callers built on the library rather than on the CLI.

    ErrUnsupported moved into the public issueops package and is aliased back from internal/storage, so a consumer holding only a role interface can classify the refusal with errors.As without importing an internal package. It is the same type throughout: beads.ErrUnsupported and storage.ErrUnsupported are unchanged aliases and every existing errors.As site keeps matching. The portable way to page a listing across backends is the keyset position (ListRequest.AfterCreatedAt/AfterID), which every implementation honors and which is stable under concurrent writes; ready work carries no keyset and therefore has no portable paging.

  • Breaking: a dependency type is now bounded at 32 characters, not 50 (bd-yby99.3). The type column on both dependencies and wisp_dependencies is VARCHAR(32) and no migration widens either, so the validator's old 50-character bound accepted a range of types — 33 to 50 characters — that no row could ever carry. Depending on the caller, such a type either failed at the database with a raw truncation error or was silently cut short. types.DependencyType.IsValid now bounds at the new exported types.MaxDependencyTypeLen, so the refusal arrives from one place, ahead of every backend, with a typed validation error.

    This narrows what bd accepts on twelve paths, among them bd create --deps, bd dep add --type, bd link --type, bd batch, bd graph apply and markdown import. In practice nothing in the wild should notice: every well-known type is far shorter (the longest, conditional-blocks, is 18 characters), and bd create --deps accepts only well-known types anyway. The user-visible error text changed too — messages that read max 50 chars or at most 50 characters now name 32 — so a script matching on that literal string needs updating. One rough edge is unchanged and worth knowing about: bd create's interactive form warns and drops an invalid dependency rather than failing, so a 33-to-50-character type there now silently loses the edge instead of creating an unstorable one.

  • Breaking: bd update --status <done-status> now enforces close policy. Moving an issue into closed (or any configured done-category status) via bd update, bd batch update, or the issueops facade now refuses when the issue has open children or a live direct blocker — matching bd close. Override with bd update --force, which now means: allow -a/--assignee to overwrite another actor's live in_progress claim, AND allow a status change into done despite open children or a live blocker (same as bd close --force). bd update --force without -a was previously a validation error on the direct path; it is now valid. In bd batch, spell the override update <id> status=closed force=true; an unforced refusal rolls back the entire batch. Facade consumers: set UpdateRequest.ForceClosePolicy. Tracker sync-pull always forces (remote state is authoritative). bd batch close remains unchecked, as before.

  • Public beads.Storage interface gained a required IssueClaimer() method (bd-serve claim role). IssueClaimer() (issueops.Claimer, error) returns the guarded atomic claim: Claim(ctx, issueops.ClaimRequest) (issueops.ClaimResult, error), one compare-and-set that sets assignee and status only while the issue is claimable, reports the idempotent re-claim as Changed: false with nothing written, and refuses a lost CAS with a wrapped ErrAlreadyClaimed or ErrNotClaimable. When the refusing state was readable in the same transaction the refusal is an *issueops.ClaimConflictError carrying the assignee and status — so a client classifies a conflict from typed fields instead of matching substrings in the message, while errors.Is and the refusal's exact prose are unchanged. POST /v0/beads/issues/{id}:claim now runs on the role; the CLI's bd update --claim is not on it yet. It is a role of its own rather than a fifth verb on issueops.Lifecycle — a capability that role does not cover gets its own accessor. Consumers that only call the interface are unaffected; any external type that implements it (a custom store, mock, or proxy) must add the method to compile. A decorator must implement it by recursing into the store it wraps, the way IssueLifecycle() does. ClaimResult.Issue is the post-state issue ROW — deliberately without labels, dependency records or comments, unlike UpdateResult — because that is what the already-published v0 claim response and bd update --claim --json both report today.

  • issueops.ReopenRequest gained a Provenance field, and bd reopen now names the issue in its Dolt commit message on the direct path. Provenance is the same optional label UpdateRequest already carries: it names the surface a mutation came from, it never changes whether a history entry is recorded, and empty still selects the implementation's default. It exists on this request because the defaults do not agree — the store-backed backends write bd: reopen issue and the unit-of-work backend writes reopen issue, neither naming the issue. A caller that spells a label now gets one entry whichever backend served it. The visible change is in bd dolt log only: a direct bd reopen bd-1 reads bd: reopen bd-1 where it read bd: reopen issue, or reopen issue on the unit-of-work backend.

    UpdateRequest and ReopenRequest are the only two requests that carry the label, and deliberately so. CreateRequest gains nothing — no surface has ever named the created issue, so the field would have no caller. Neither do CloseRequest, CloseBatchRequest, ClaimNextRequest, AddCommentRequest, AddDependenciesRequest or RemoveDependencyRequest: each of those roles composes its entry from what the call actually did, so a caller-supplied label could only restate it or contradict it. For the batch close and the claim one is not even expressible — those messages name what LANDED and which id the claim WON, neither of which is knowable before the call. For the dependency verbs the entry follows the edges: one edge names both endpoints, several name only their count, and — see the source-routing note below — a request made entirely of wisp-sourced edges records no entry at all, leaving a label nothing to name.

  • Public beads.Storage interface gained required DependencyEditor(), Commenter() and IssueRelations() methods. Three more roles alongside IssueLifecycle(), IssueReader(), ReadyClaimer() and BatchCloser(), reached the same way and layered by each decorator in its own accessor. Consumers that only call the interface are unaffected; any external type that implements it (a custom store, mock, or proxy) must add all three to compile, and a decorator must implement the two write roles by recursing into the store it wraps rather than delegating blindly.

    DependencyEditor() (issueops.DependencyEditor, error) edits the dependency graph. AddDependencies asserts a set of edges as ONE durable act — one transaction, at most one history entry — and is ALL-OR-NOTHING: any refused edge (self-dependency, cycle, type conflict, hierarchy conflict, unknown endpoint) refuses the whole request and writes nothing. Edges asserted together describe a graph, and half a graph is a graph nobody asked for. RemoveDependency is idempotent: a missing edge is Removed: false with a nil error and no history entry, not ErrNotFound.

    The dependency-type constants issueops.DepBlocks, DepParentChild, DepRelated, DepDiscoveredFrom, DepRepliesTo and DepWaitsFor ship with it. They are documented common values of an OPEN, workspace-configurable set — not a closed enum, not exhaustive, and not validated against: nothing refuses a type merely because it is absent from the list, and a workspace's own type is a first-class value everywhere a DependencyType is accepted. They exist because internal/types is unimportable outside the module, so without them an external embedder can hold a DependencyType but has no spell-checked way to name one — the same reason the error sentinels are re-exported.

    Commenter() (issueops.Commenter, error) appends one comment as one atomic mutation. A blank Text is ErrValidation, as is an empty IssueID; the issue-then-wisp fallback happens inside, as it does for Reader.Get; a non-empty id naming neither plane is ErrNotFound — an unaddressed request is invalid, not a thread that went missing, so the two are worth telling apart. issueops.Comment (an alias for the canonical comment model) is the result type. A comment on an ephemeral row records no durable history entry — the wisp tables are dolt-ignored so ephemeral work never ships, and an entry naming a wisp thread is exactly the artifact that ignoring them prevents; the comment still lands on the ephemeral thread and reads back from it.

    IssueRelations() (issueops.Relations, error) reads an issue's neighbors. Related requires an explicit RelationOut or RelationIn — the zero direction is ErrValidation, with no implicit "both", because a silent default hands a caller the inverse graph with nothing to notice. An unusable Types entry is ErrValidation rather than a filter that matches nothing, and an anchor missing from both planes is ErrNotFound rather than an empty list. The result order is pinned: ascending by the neighbor's id, with the edge type breaking a tie.

    bd dep add, bd dep remove, bd dep list, bd comment and bd comments add are on these roles on BOTH routes — the direct one reaches them through the store's own accessor exactly as the proxied one reaches them through the provider's. Four behaviors changed on the proxied route: bd dep list now answers in the pinned order rather than in the order the two dependency tables happened to be read; a bd dep list naming an issue that does not exist now reports that instead of printing an empty neighbor list, matching what the direct route has always done; a bd dep remove that finds no edge now records no Dolt commit at all, where it used to name one; and a single-line bd dep add --file writes bd: dep add <a> <b> rather than dependency: add 1 edges.

    Four more changed on the direct route as it joined them. bd dep list there now answers in the same pinned order. Dolt commit messages for the explicit dep verbs are now the role's — bd: dep add <a> <b> and bd: dep remove <a> <b> — where the direct path used to write dependency: add <type> <a> -> <b> / dependency: remove <a> -> <b> in server mode and a generic bd: dep add (auto-commit) by <actor> [...] in embedded mode. A bd dep add --file edge refused by storage no longer carries a line N: prefix, because the role's request is all-or-nothing and names no edge index — parse and id-resolution errors still name their line. And a bd dep add --file in embedded batch/off auto-commit mode now defers its version commit to the next explicit commit point instead of always writing one, which is what every other write verb already did.

    bd comment's direct route changed one thing with it, on the embedded backend only: a comment's Dolt commit is now the role's bd: comment <id> authored by beads, where the CLI auto-commit wrote bd: comment (auto-commit) by <actor> [<id>] authored by root. That is the spelling the Dolt server backend already wrote, and it is one commit per comment either way. Comment LISTING (bd comments <id>) is on neither role and unchanged on both routes: reading a thread is a paging question with a cursor of its own, which issueops.Commenter deliberately excludes.

    DependencyEditor.AddDependencies routes each edge by its SOURCE, the rule the stores' own dependency verb has always applied and the one bd dep remove and bd dep list already read: a wisp-sourced edge goes to wisp_dependencies with its event in wisp_events, an issue-sourced one to dependencies and events. Routing by source rather than pinning it to the issues plane is what keeps bd dep add <wisp-id> <target> working: a wisp has no row in the issues plane, so a pinned source has nothing to hang an edge off, and the two routes would not even have failed alike — the direct one on its source-existence check, the proxied one on an fk_dep_issue foreign key violation. ONE REQUEST MAY MIX THE TWO: it is still one transaction, the parent-child-first ordering and the whole-graph cycle gate both span the two tables (a cross-plane cycle is still refused), and a refusal rolls back both planes. Only the durable half is versioned, so a request made entirely of wisp-sourced edges records no history entry and leaves the working set clean — asserted end to end against a real embedded store by TestEmbeddedDepAddWispSource, which runs bd dep add <wisp-id> <target> at the front door. bd link on the proxied server routes the same way, matching its direct twin. Behind the role there is no plane-pinned bulk variant left to reach for by mistake: the internal domain layer's three bulk add verbs are now the one source-routed AddDependencies, so every bulk caller — role, bd link, unit of work — gets this rule and cannot opt out of it.

  • Public beads.Storage interface gained required ReadyClaimer() and BatchCloser() methods. Two more roles alongside IssueLifecycle() and IssueReader(), reached the same way and layered by each decorator in its own accessor. Consumers that only call the interface are unaffected; any external type that implements it (a custom store, mock, or proxy) must add both methods to compile, and a decorator must implement them by recursing into the store it wraps rather than delegating blindly.

    ReadyClaimer() (issueops.ReadyClaimer, error) takes ready work atomically: ClaimNext selects, wins the compare-and-set and hydrates the row it won, all in one transaction. Its filter is issueops.ReadyRequest — the same type Reader.Ready takes — so the claim's predicate cannot drift from the set bd ready shows; Limit and Offset on it are ErrValidation, because the claim scan clears them and accepting a field only to drop it is worse than refusing it. An empty ready front is a nil result with a nil error, not an error class to classify.

    BatchCloser() (issueops.BatchCloser, error) closes many issues as one durable act: CloseBatch commits every item that lands in ONE transaction with at most one history entry. The request is the transaction boundary — nothing on the contract exposes begin or commit. It is deliberately not all-or-nothing: a refused id is skipped and the survivors commit together, so CloseBatchResult.Outcomes carries one entry per requested item in request order, each with a typed errors.Is-able refusal that never aborts the rest. An optional ClaimNext runs after the closes, in the same transaction, only when at least one item closed.

    bd ready --claim and bd close are on these roles on BOTH routes. Three behaviors changed on the proxied route: the proxied claim's Dolt commit message is now bd: claim ready <id>, the spelling the other two backends already wrote (it was bd: ready --claim <id>); molecule auto-close, --suggest-next and --continue now run after the close transaction rather than inside it, which is where the direct route has always run them — a plain bd close a b c is still exactly one Dolt commit, while a close that actually auto-closes a molecule or advances a step now produces two; and bd close --claim-next walks the ready order and takes the first row it can win, where it used to read the single top ready row and warn if that row had already been taken — a race with another agent now claims the next candidate instead of claiming nothing.

    bd close's direct route closed N ids as N transactions with N Dolt commits until it joined them, so five things changed as it did. bd close a b c now writes ONE commit named bd: close a, b, c instead of three named bd: close issue — one request per STORE, since a transaction cannot span the second repo a prefix-routed id reaches. --claim-next runs inside the closes' transaction and only when something landed, walks the ready order the way the proxied route already did, and its --json claimed value is now the post-claim row (in_progress, assigned) rather than the pre-claim row the old two-step read. Molecule auto-close runs after ALL the closes land rather than between them — the root still closes exactly once; only which id's slot prints the Auto-closed completed molecule line can move. The open-children refusal moved from a cmd/bd pre-flight read into the close's own transaction, closing a read-then-write window that made bd close <child> <parent> depend on argument order; its message is unchanged. And an unsatisfied gate on an issue that ALSO has open children now reports the gate rather than the children, because the gate check stayed in the pre-flight — both were already refusals.

  • Public beads.Storage interface gained a required IssueReader() method (bd-serve reader role). IssueReader() (issueops.Reader, error) is the read counterpart of IssueLifecycle(): one accessor returning a role that answers Ready, List and Get with filter construction, workspace-config loading, default limits and the wisp fallback all performed inside it, so a caller cannot half-perform that construction and answer the same question a different way. All three of bd serve's issue reads and bd show --json's detail view are on the role; bd list and bd ready are not yet, and share the internal/workapi builders instead — see issueops.Reader's doc comment for exactly what that does and does not cover. It is a role of its own rather than three more methods on issueops.Lifecycle — a capability that role does not cover gets its own accessor. Consumers that only call the interface are unaffected; any external type that implements it (a custom store, mock, or proxy) must add the method to compile. A decorator must implement it by recursing into the store it wraps and layering its own behavior onto the result, the way IssueLifecycle() does, rather than delegating blindly — a blind delegation compiles and still satisfies the interface while silently dropping that decorator's layer for every read.

  • beads.BulkIssueStore.ReclaimExpiredLeases gained a types.ReclaimFilter parameter (wy-jpd3.3), threaded through the domain use-case/repository interfaces and every backend. Callers that reclaim globally pass the zero types.ReclaimFilter{} (unchanged behavior); any external type that implements the interface must update its signature to compile.

  • Public beads.Storage interface gained two required methods (#4911). UpdateIssueChecked (an optional ExpectedVersion compare-and-swap on updates) and MergeMetadata (atomic single-key metadata merge) are now part of the beads.Storage / storage.Storage contract. Consumers that only call the interface are unaffected; any external type that implements it (a custom store, mock, or proxy) must add both methods to compile. UpdateIssueChecked with a nil ExpectedVersion behaves exactly like UpdateIssue, and MergeMetadata can be layered over a read-modify-write of issue metadata. See examples/library-usage.

  • Storage backend scope simplified (bd-sadcd). The recently merged direct PostgreSQL and MySQL adapters have been rolled back before entering a tagged release. Supporting additional general-purpose server databases introduced dialect, credential, schema-lifecycle, migration, CI, and operational complexity at odds with our goal of keeping Beads as simple as possible and consuming as few resources as possible. The storage interface, shared issue core, SQLite implementation, and conformance harness remain; embedded Dolt, Dolt server, and SQLite are the supported storage paths. Dolt server mode and its use of the MySQL wire protocol are unchanged. Existing PostgreSQL or MySQL workspaces stop before their configured databases are opened or modified; see the migration guide.

  • Cross-type blocking dependencies are now allowed (bd-wg7ve, #4034). bd dep add <task> <epic> — gating a work item on an epic (program) completing — previously failed with a backwards-reading error ("tasks can only block other tasks, not epics"). The blanket same-type rule (GH#1495) is replaced by a hierarchy deadlock guard that rejects only the cases that actually wedge the graph: gating an issue on its own ancestor (the ancestor can't close until its descendants finish) or on its own descendant (blocked status cascades down to the very issue that must close to clear the gate). Sibling ordering edges stay allowed. The guard now also covers conditional-blocks, which previously skipped cross-type validation entirely. A task gated on an epic becomes ready when the epic itself is closed, consistent with bd ready/bd blocked.

  • Claim/unclaim refusals now steer toward the holder instead of teaching eviction (bd-at6rc, wyvern wy-zs5s2). The --claim refusal for an issue assigned to someone else suggested bd unclaim <id> — in a multi-agent fleet that copy taught an unclaim-then-claim steamroller that evicted a live, heartbeated claim mid-review. The refusal now says to coordinate with the holder, and unclaim's ownership rejection frames --force as an abandoned-claim escape hatch rather than a routine override. (The ownership check itself — foreign unclaim requires --force — landed in #4675.)

Fixed

  • Multi-id bd dep list no longer changes its JSON shape on failure (bd-nhsno). The direct route used the batched edge read only if err == nil and otherwise fell through to the neighbour path, which emits a different document — issues, not edge rows — with no error and no warning. Worse, the batched read was only ATTEMPTED when every resolved anchor shared one store, so a cross-workspace batch emitted the wrong document every time rather than only on failure. The read now goes through the new issueops.EdgeReader role, taken per store, with the answers merged back into argument order, and a failure is reported as a failure.

  • bd ready --claim honours the directory-label scope (bd-s10oa). The branch that applies a directory's configured directory.labels default tested the filter it had already written that default into, so it could never fire. In a directory with directory.labels configured, bd ready --claim therefore claimed from the WHOLE ready front while bd ready beside it listed only the configured scope — handing an agent work the listing never offered it. Both routes affected.

  • Proxied bd config set status.custom and types.custom take effect immediately (bd-kzepq). Reads of the custom status and type vocabularies are TABLE-first: they consult the normalized custom_statuses and custom_types tables and fall back to the config string only when the table is empty. The direct route has always rewritten the table in the same transaction as the row; the proxied route wrote only the string. So on a team server bd config set types.custom "session" reported success, bd config get types.custom read it back, bd doctor re-verified against the string and reported all-OK — and bd create -t session answered invalid issue type: session for as long as the workspace lived. Both routes now project the value into its table inside the same unit of work as the row, so the setting and its effect land together or neither does.

  • bd config set-many issue_prefix=<x> is refused, as bd config set has always been (bd-kzepq). bd config set rejects the workspace id prefix in both spellings and points at bd init --prefix, bd bootstrap and bd rename-prefix, each of which does work a config write cannot — seeding a workspace that has none, or rewriting the ids that already exist. set-many partitioned its pairs by source and wrote the database ones straight through, so it walked past that guard and re-prefixed the workspace: beads created before the write and beads created after it disagreed about their own namespace, with nothing to reconcile them. Both verbs now refuse it, and so does the substrate, so a future front door cannot miss it either.

  • bd create --id <id> on a proxied server refuses an occupied id instead of overwriting it (bd-7oyh5). The direct route has always been create-only. The proxied route passed no such flag, so bd create --id bd-abc123 --title ... against an id that already existed silently upserted every column of the resident issue — title, type, description, labels — and exited zero. The original was gone, with no history entry naming a create and nothing in the output to suggest anything had been replaced. Both routes now return ErrAlreadyExists naming the id and leave the stored row untouched, across the shared id space: an occupied durable id refuses a wisp create and an occupied wisp id refuses a durable one. If you have automation that relied on --id as an upsert, it should call bd update.

  • Proxied bd update --ephemeral and --no-history move the issue instead of flagging it in place (bd-xt6de). The direct route sends these through an atomic aggregate move: insert into the target plane, copy the auxiliary rows, retarget inbound dependency edges, drop the lease, delete from the source. The proxied route instead ran a plain column update on whichever table the row already sat in, so a proxied bd update --ephemeral left the issue in issues carrying ephemeral = 1 — still durable, still versioned, still replicated, but invisible to every wisp-plane read and skipped by JSONL export. Both routes now perform the move. The inverse spellings --persistent and --history appeared to work before only by accident: moving a row that never left issues is a no-op that still clears the flag. This does not repair existing rows. Any issue a proxied bd update --ephemeral produced before this release is still in issues with the flag set, and nothing here finds it; that needs its own migration.

  • Proxied bd reopen reopens an issue in a configured done status (bd-1f5a2). The proxied handler short-circuited on status != closed and reported "nothing to do" for any workspace-configured done status, so on a team server an issue in a custom done state such as verified could not be reopened at all while the same command worked locally. Both routes now go through issueops.Lifecycle.Reopen, which spans the whole configured done category. Reopening something genuinely not closed is still a no-op and still says so.

  • bd update --parent on a proxied server no longer leaves a stale parent (bd-yby99.26). A parent id is a set assignment: setting it replaces every parent edge the child has, and clearing it removes them all. The unit-of-work route stopped after the first edge it found. A child with two parents reparented to a third kept one of the originals, and clearing the parent left the rest in place — in both cases reported as a success. Two parents on one child is not the common shape, but the failure was silent and the resulting graph is wrong for every reader that walks parent edges. The Dolt backends already applied the set rule; this brings the third route into line.

  • bd dep remove on a wisp-sourced edge now removes it (bd-yby99.17). On the proxied server the removal was pinned to the durable plane: an edge whose source is a wisp was looked for in dependencies, not found, and reported as Removed: false with a nil error while the row stayed in wisp_dependencies. The command exited zero having changed nothing. A removal cannot put an edge anywhere, so it now reads the source's plane and deletes from there, and the delete's own result is the found/not-found verdict. Removing an edge that genuinely does not exist is still a nil-error no-op, as before.

  • A dependency on an issue in another repository no longer fails with a raw foreign-key error (bd-ocrn7). beads addresses a foreign repo's work by prefix — otherrig-9001 from a rig whose own prefix is bd- — and the database holds such a target in depends_on_external, a column with no foreign key, precisely because the row cannot exist locally. The domain route classified the target by looking only for the literal external: prefix, decided otherrig-9001 was local, wrote it to depends_on_issue_id, and the foreign key rejected the insert with Error 1452. No edge was written and the error named a constraint rather than the problem.

    Both routes now classify through one named rule, issueops.IsExternalDepTarget: a target is external if it is spelled external:... or its prefix differs from the source's. That is what the Dolt backends have always done, so this brings the third route into line rather than inventing a policy. The foreign key itself is untouched, and a genuinely local, genuinely missing target is still refused as before. One consequence beyond the failing case: a cross-prefix target that does happen to exist in the local issues table now also lands in the external column on this route, which means derived blocking no longer reads that edge — again matching what the Dolt backends already produced.

  • A ready claim that wins an ephemeral row no longer reports a spurious "could not be verified" error (bd-yby99.4). In server mode, bd re-reads a claim it just made to confirm the write landed, because an exit code is not truth against a degraded server. That re-read always looked in the issues table. But a ready claim can win a wisp when the request's filter admits ephemeral rows (bd ready --claim --include-ephemeral), and the claim is written to the wisps table by design — so the verify hunted for the row in a table that never held it and turned a perfectly good claim into reported success but could not be verified (server degraded?). The verification is now plane-aware rather than skipped: a wisp claim is checked against the wisps row, so no check was traded away for the fix.

    The ClaimNext contract is now explicit that the ready set is whatever the filter admits — an ephemeral row the filter pulls in is genuinely claimable, and claiming it moves that row rather than promoting it to a durable issue. Two properties follow that callers should plan around: such a claim records no durable history entry, because ephemeral rows are not versioned, so reconstructing who took what from history alone will not show it; and it takes no lease, so bd heartbeat refuses it and bd reclaim will not recover it if the agent holding it dies.

  • bd dolt push and bd sync no longer adopt a Dolt remote derived from git origin without consent (#5068). On a rig with no Dolt remote configured, both commands silently derived one from git remote get-url origin, added it, persisted sync.remote into .beads/config.yaml, committed that config change under the user's git identity, and uploaded the full issue history — no prompt, no flag, no opt-out. A public git origin therefore published the whole issue database on a command the user believed targeted an already-configured remote.

    Adoption is now a consent decision and fails closed. Interactively, bd shows the derived URL and every side effect that follows a yes (including the config commit made under your git identity) and defaults to no. Non-interactively it refuses and exits non-zero, naming the URL it would have adopted and the explicit opt-in. --yes/-y consents ahead of time for scripted use; --no-adopt or BD_NO_REMOTE_ADOPT=1 disables adoption entirely and wins over --yes. Rigs with a remote already configured are unaffected — that path was and remains a no-op.

    Workspace resolution also moved below the gate: it calls prepareSelectedNoDBContext, which mutates workspace state, and nothing may mutate before consent is established.

  • Proxied-server CI shard 1 flake: TestProxiedServerCleanDatabases ran a server-global destructive command against the shared test container (p1-9lf, hazard tracked in p1-8dz). The test used the shared external Dolt container under t.Parallel(), and bd dolt clean-databases drops every database matching staleDatabasePrefixes on the server it is pointed at. The container is bootstrapped with a database named beads_test (internal/testutil/container_provider.go), which matches one of those prefixes — so every run of this test dropped the shared container's own bootstrap database out from under whatever sibling tests were mid-flight, which is timing-correlated with the busy buffer / "pending schema migrations alter pre-existing dirty tables" failures seen in TestProxiedServerDelete and TestProxiedServerFindDuplicates. The test now runs against a dedicated proxied server (bdProxiedInit), like the new purge test, so its blast radius is its own server.

  • bd dolt clean-databases --purge-dropped now works in proxied-server mode (p1-9lf). The command read the flag and then dropped it on the floor before dispatching to the proxied-server implementation, so under --proxied-server stale databases were dropped but never purged: their directories stayed under .dolt_dropped_databases/, disk was never reclaimed, and nothing reported that the flag had done nothing. The proxied dual had drifted further than that one flag — it also lacked the batching/circuit-breaker drop loop, the dry-run "purge ignored" notice, and the DOLT_UNDROP recoverability trailer, and its early return on "nothing stale found" was incompatible with purge semantics (a prior run's residue is invisible to SHOW DATABASES, so --purge-dropped must still fire when this run drops nothing). Both topologies now share one implementation keyed on versioncontrolops.DBConn, which *sql.DB (direct server) and the pinned non-transactional *sql.Conn (proxied server, via uow.MaintenanceProvider.RunNonTx) both satisfy, so the behavior cannot diverge again. Per-operation timeouts now derive from the command's context on both paths, so Ctrl-C interrupts a long clean.

  • bd dolt clean-databases gains an opt-in --purge-dropped flag to reclaim disk from what it drops (be-pq5, #3663). DROP DATABASE only moves a database's directory under .dolt_dropped_databases/; Dolt keeps the on-disk data there — recoverable via CALL DOLT_UNDROP(name) — until an explicit CALL DOLT_PURGE_DROPPED_DATABASES(). The operator-facing cleanup command dropped stale testdb_*/benchdb_*/etc. databases from SHOW DATABASES but never issued that purge, so disk usage on the shared Dolt server stayed high across repeated runs. --purge-dropped runs the purge after cleanup (a failed purge is reported as a warning rather than aborting the command, since the databases are already dropped from the server's active set at that point) and fires even on a run that finds nothing stale to drop, since a prior run's dropped-but-unpurged residue isn't visible to SHOW DATABASES either. CALL DOLT_PURGE_DROPPED_DATABASES() is server-global and irreversible — Dolt has no way to scope it to only the databases a given run dropped, so --purge-dropped also permanently deletes every other dropped-but-not-yet- purged database on the server, removing DOLT_UNDROP recovery for all of them. It defaults to off for that reason; without it, dropped databases are left in place, recoverable via DOLT_UNDROP, same as before this change.

Added

  • Conditional (compare-and-set) updates: bd update --if-assignee / --if-status (bd-wsqvw, from wyvern's wheelhouse hostile-review epic wy-mdi5h; design in PROPOSAL-cas-conditional-update.md). When either guard is present the update applies only if the issue's current assignee/status still equals the expected value — one atomic transaction, nothing written on a mismatch, and a loud non-zero exit naming actual vs expected (typed as storage.ErrAssigneeMismatch / new ErrStatusMismatch). --if-assignee '' means "expected unassigned". A guard mismatch is machine-distinguishable from infra failure: exit code 13 when every failure in the run was a stale guard (a racer won — skip gracefully) vs 1 for anything else (retry/abort); in --json mode each entry in the failure report's failed array additionally carries "guard_mismatch": true, and the stderr text always contains the sentinel token assignee mismatch / status mismatch. This closes the two coordination transitions no existing verb could express: reassign X→Y only while X still holds it (bd update <id> --if-assignee worker -a mayor), and claim-on-behalf with a status guard (bd update <id> --if-assignee '' --if-status open -a owner -s in_progress). Guards require a field update to ride on, are mutually exclusive with --claim (its own CAS), compose with each other and with the engine's ExpectedVersion row CAS, and work in every dolt mode. In Dolt server mode a guarded update that writes assignee/status is claim-family and is resolved by the bd-zccb9 verify-by-re-read protocol, so its exit code stays truthful under a degraded server. Library consumers get the same guards via new UpdateIssueOptions.ExpectedAssignee/ExpectedStatus fields on the existing UpdateIssueChecked — no interface change; out-of-tree Storage implementations that ignore the new fields simply do not enforce them and should add support before advertising guard semantics.

  • Pool-aware claiming via the claim.pools config key (bd-bguz6). Dispatcher fleets pre-assign issues to a pool pseudo-assignee (e.g. fable-crew); --claim previously refused those ("already assigned"), forcing a two-step --assignee <me> -s in_progress per issue. Aliases listed in bd config set claim.pools "fable-crew,night-crew" are now claimable by any actor through the same atomic compare-and-swap — the claim stamps the normal lease, and issues assigned to a real actor (or to an alias not in the config) keep their anti-steal protection. Pool-aware claiming works in every dolt mode — the proxied-server claim path applies the same predicate — and claim.* is a recognized config namespace (documented in bd config --help). Note: if a pool take's lease expires, bd reclaim returns the issue to the unassigned pool, not to the pool alias it was dispatched to. Off by default: with no claim.pools configured, behavior is unchanged.

  • Work leases: claim-TTL, heartbeat, and reclaim for dead-worker recovery (schema v54, migration 0054) (#4537). A claim was previously permanent — a worker that died mid-task stranded its issue in_progress forever. Claims now carry a lease:

    • Claiming stamps lease_expires_at = now + TTL (default 5m) and heartbeat_at.
    • bd heartbeat <id> — owner-only; pushes the lease forward. Fails once the lease is gone.
    • bd reclaim --older-than <dur> — reverts in_progress issues whose lease expired more than <dur> ago back to ready (clears assignee/started_at, records a lease_reclaimed event). Default grace 2×TTL.

    Because Dolt has no row locking and merges concurrent commits cell-by-cell, every status/ownership/lease-mutating path also rewrites a shared row_lock cell, forcing a racing heartbeat vs. reclaim to a serialization conflict that the retry layer replays — instead of silently cell-merging into a zombie claim. The same landing wraps the work-queue hot paths (bd ready --claim, claim, update, close) in serialization- conflict retry, so N concurrent workers draining one queue no longer surface MySQL 1213/1205 errors.

  • bd migrate --force (and bd migrate schema --force): CLI flag twin of BD_ALLOW_REMOTE_MIGRATE=1 for bypassing the remote-migrate gate (#4259) as the single designated migrator; process-local so it cannot leak into child processes (git hooks, dolt subprocesses).

  • Cursor agent hooks. bd setup cursor now installs .cursor/hooks.json alongside the existing rules file, wiring three Cursor lifecycle events to a new hidden bd cursor-hook command:

    • sessionStart injects full bd prime context into every new agent session.
    • preCompact arms a one-shot refresh marker (and notifies the user).
    • postToolUse re-injects bd prime exactly once after a compaction, then no-ops.

    This brings Cursor (IDE and recent cursor-agent CLI builds — verified on the 2026.06 line; early-2026 CLI builds only fired shell hooks) to parity with the Claude Code / Codex hook integrations, so Beads context survives context compaction instead of being forgotten. Existing user hooks in .cursor/hooks.json are preserved, and bd setup cursor --remove only removes the Beads-managed entries.

    The integration now matches Claude/Codex on several more fronts:

    • bd init auto-installs Cursor (rules + skill + hooks) the same way it auto-sets up Claude Code and Codex, and stages .cursor/ for commit.
    • bd setup cursor --global writes hooks to ~/.cursor/hooks.json (and the agent skill to ~/.agents/skills/beads) so they apply to every project. (Global scope is hooks + skill only; Cursor has no reliable file-based global rules location — those live in Cursor Settings.)
    • Canonical rules content. .cursor/rules/beads.mdc now wraps the shared recipes.Template (the same content every other file-based recipe uses) instead of a hand-maintained copy, so it no longer drifts or omits the Issue Types / Priorities / git-authority sections.
    • Agent skill. bd setup cursor installs the shared .agents/skills/beads/SKILL.md (which Cursor loads natively), so Cursor-only users get the same progressive-disclosure skill Codex installs. The skill is shared with Codex and is only removed once no integration still relies on it.
    • bd doctor reports Cursor Integration, Cursor Settings Health (malformed .cursor/hooks.json → error), and Cursor Hook Completeness (all three lifecycle events present) checks — with agent-mode enrichment — paralleling the Claude integration checks.

Changed

  • Remote-ahead adopts fast-forward automatically when it is provably loss-free AND lands exactly at this binary's latest migration. When the smart remote-migrate gate (#4516) finds the remote ahead with no content skew, this clone's local Dolt history a strict ancestor of the remote's with a clean working set, AND the fast-forward would land exactly at the binary's own latest migration (nothing pending after it, nothing beyond what this binary supports), bd now fast-forwards to the remote's migrated schema automatically instead of stopping with an adopt directive — nothing local is discarded. An unpushed local commit, a dirty working set, or a remote that is not exactly at this binary's latest migration all disqualify the automatic fast-forward and fall back to the existing manual bd bootstrap adopt directive (with the sharper loss-free guidance whenever ancestor+clean still hold), never a forced write. Set BD_SMART_GATE=0 to opt out and keep the manual wall for every remote-ahead case, as before (#4259).

Fixed

  • Claim-family writes are now verified by re-read in Dolt server mode (bd-zccb9, from wyvern incident wy-ejph3). Under a degraded sql-server the exit status of bd update --claim / bd claim / unclaim was not truth in either direction: a claim could report success while the server-side transaction died with the abandoned connection and rolled back (a phantom claim that later cost a duplicate implementation), and conversely a connection error could print with the write actually applied. After the claim transaction, bd now re-reads the issue's assignee and status on a fresh connection and resolves the outcome against the database: a reported success that did not land fails loudly ("treat the claim as NOT applied"); an ambiguous commit-phase loss is settled by the re-read — verified applied becomes an accurate success, verified rolled back is replayed once (safe: nothing landed). Applies to claim, ready-claim, and unclaim in server mode; wisps and embedded mode are unchanged. New metrics bd.claim_verify_lost_total and bd.claim_verify_recovered_total count loud failures and converted outcomes.

  • Single-issue bd create --id P.N no longer leaves child_counters stale (one root cause of #4750; the issue stays open — three other drift paths remain). CreateIssueInTxWithResult only reconciled child_counters when the batch path called ReconcileChildCounters; an explicit hierarchical --id never advanced the parent's last_child high-water mark. Once such children were later archived out of issues, GetNextChildIDTx no longer saw them and could re-mint an already-used suffix — an ID collision with an archived issue. The single-issue path now calls ReconcileChildCounters too, and its changed tables are merged into the result so the reconcile actually gets staged for commit.

[1.1.2] - 2026-07-26

Hotfix release cut from v1.1.0. (There is no 1.1.1 release: its tag was cut but the release pipeline's package gate failed on a stale MCP lockfile before publishing anything, and the version number was burned rather than moving the tag.) If bd migrate on 1.1.0 aborted with rekey aux row ids: <table>: ... invalid hash length and the database then refused to open, this release lets the migration complete.

Fixed

  • The v53 aux row re-key survives dolt#11131 encoding drift. On storage affected by the upstream Dolt adaptive-encoding bug, reading the drifted cells panics (invalid hash length: 19), which aborted the migration and left the database unopenable under 1.1.0. The re-key now skips such a table with a warning, records it in a clone-local drift record (aux_row_rekey_drifted in local_metadata), and completes the migration; recorded tables are retried on later migration passes and are exempted from the changed-signature dirty-table guard so the retry cannot re-brick the database (#4380).

[1.1.0] - 2026-07-04

First stable release of the 1.1.0 line. It consolidates everything from [1.1.0-rc.1] and [1.1.0-rc.2] (see those sections below for the full feature set) plus the post-rc.2 migration-recovery fixes listed here. The theme of 1.1.0 is a safe schema-migration and upgrade path: it repairs the v52/v53 drift classes that broke real-world rc.1/rc.2 upgrades, makes an interrupted migration recoverable in-tool instead of a dead end, and turns the remote-migrate gate from a blunt block into a state-aware one.

Upgrade Notes

  • Back up before migrating (bd export --all -o backup.jsonl) before running bd migrate on an older, especially remote-backed, database. The fixes below repair databases that reached a migration with drifted state; an export is cheap insurance while such drift is being repaired.

Changed

  • The state-aware remote-migrate gate is now on by default. rc.2 shipped the smart gate behind an opt-in env var, but nothing in the gate's block message or docs surfaced that the var existed, so in practice everyone kept hitting the blunt always-block behavior. The provably-safe first-mover case (remote at the same schema version as this clone) now auto-migrates without BD_ALLOW_REMOTE_MIGRATE; remote-ahead still stops with an adopt directive and content skew still stops for a human. Set BD_SMART_GATE=0 to opt out and restore the unconditional block (#4516).

Fixed

  • A failed v53 migration no longer traps the database, and the v53 repair now covers wisp_dependencies split-column drift. rc.2 repaired the issues rig columns, but a database coming from schema v49 could still fail v53 when its wisp_dependencies table was missing the split target columns (depends_on_issue_id, depends_on_wisp_id, depends_on_external) — and a half-applied v53 then bricked every subsequent command on open (#4555). The migration runner now adds and backfills the missing split columns before v53, and a narrow recovery gate lets an already-failed-v53 database self-heal on the next open — including the issue_snapshots / compaction_snapshots tables that migration 0051 leaves dirty during a single-pass v49→v53 upgrade (#4558).
  • A dirty working set no longer deadlocks migration recovery in embedded mode. Every store open runs the migration, which correctly refuses to alter tables that have uncommitted working-set changes — but the documented recovery (commit the working set) is itself a command that opens the store and hits the same refusal, a deadlock with no in-tool escape (#4566). bd dolt commit and bd vc commit now open past that guard, commit at the current schema, after which a normal bd migrate applies cleanly; the refusal message now names the recovery path (#4567).

[1.1.0-rc.2] - 2026-07-02

Second release candidate for 1.1.0. Fixes the two upgrade-breaking migration regressions reported against rc.1 (#4502, #4534), hardens the remote-migrate gate that rc.1 introduced, and ships the validated upgrade documentation.

Upgrade Notes

  • Back up before migrating. The upgrade guide's remote-backed recipes now start with a JSONL export (bd export --all -o ...) before BD_ALLOW_REMOTE_MIGRATE=1 bd migrate. The two fixes below repair real databases that reached a migration with drifted state; an export is the cheap insurance while such drift is being repaired.

Fixed

  • v53 migration no longer fails on pre-rig issues tables. The rig/agent columns (hook_bead, role_bead, agent_state, last_activity, role_type, rig) were only ever added to the squashed bootstrap 0001_create_issues, so a database bootstrapped before they existed sits at schema v52 without them — and migration 0053, which copies those columns from wisps, failed with Unknown column 'agent_state' in 'issues' even with zero rig wisps to repair (#4502). The migration runner now repairs the drift in code immediately before applying v53, adding whichever of the six columns are missing (databases in the wild may have some but not all). Shipped migration files stay frozen — the repair lives in the runner because a failing migration can never be fixed forward by a later migration file.
  • One orphaned child_counters row no longer bricks every bd create. Migration 0039 dropped fk_counter_parent and clone-local migration 0002 re-added it under FOREIGN_KEY_CHECKS = 0, so a counter row orphaned during the FK-less window (an interrupted create, a parent deleted without cascade) survived the constraint's return — and Dolt then failed constraint validation on every subsequent insert, including brand-new top-level issues, on an otherwise healthy database (#4534). A new clone-local migration moves any live-wisp counters to wisp_child_counters and deletes rows dangling from issues — the same rows the FK's ON DELETE CASCADE would have removed had it been in force. Runs automatically on the next command with the new binary; already-bitten databases are healed in place.
  • Old binaries fail fast on writable opens of a schema-newer database instead of proceeding against a schema they do not understand (#4531) — the guard rail for the mixed-version window during multi-clone upgrades.
  • Prerelease GitHub releases now ship prerelease-correct install instructions. The release notes header previously showed the stable install methods (brew, install scripts), three of which do not deliver a prerelease (#4530).
  • bd remember no longer clobbers a memory whose content is its own key, and a bare bd remember <existing-key> now recalls instead of overwriting.

Added

  • Smarter remote-migrate gate. The gate introduced in rc.1 is now state-aware and agent-safe: cases that are provably safe to migrate auto-resolve instead of stopping every agent at the wall, while genuinely risky states still require the designated migrator (#4515, #4516).
  • Backend-agnostic storage conformance test suite (#4414).

Documentation

  • Validated upgrade recipe for remote-backed / multi-clone databases in the upgrade guide, exercised end-to-end on a live database (#4514), plus a consolidated Homebrew tap-migration snippet across install docs, a README pointer to the upgrade guide, and a pre-migrate backup step in the recipes (this release).

[1.1.0-rc.1] - 2026-06-23

Upgrade Notes

  • Mixed bd versions sharing one Dolt remote. Pre-1.0.6 binaries record applied migrations as (version, NULL) while 1.0.6+ records (version, sha256) in schema_migrations, so two clones applying the same migration with different bd vintages used to produce a row conflict on bd dolt pull. The pull auto-resolver now resolves this class by keeping whichever side recorded the hash. Two different recorded hashes for the same version are the real #4259 schema fork and are still surfaced to the operator (see the bd doctor Migration Content Skew check). To avoid the mixed-vintage window entirely, upgrade all clones of a shared remote together, letting one designated machine migrate and bd dolt push first.

  • Upgrading a remote-backed database: one machine migrates, the rest adopt. This release reshapes the dependencies primary key (migration 0050), and bd now refuses to silently migrate a database that has a Dolt remote configured. If two clones cross the 0050 boundary with un-synced dependency edits on both sides, their histories become permanently un-mergeable — Dolt refuses the merge outright (cannot merge because table dependencies has different primary keys in its common ancestor) before the pull auto-resolver can run. Replacing the binary alone is not enough; follow the ordered recipe:

    1. With your current binary, on every clone: bd dolt push + bd dolt pull until all are in sync, then stop editing. Once the new binary is installed, push/pull are gated too, so this must happen first.
    2. Designated migrator only: install the new binary, then BD_ALLOW_REMOTE_MIGRATE=1 bd migrate and bd dolt push.
    3. Every other clone: install the new binary, then bd bootstrap to adopt the migrated database (bd dolt pull is refused while the clone still has pending migrations; do not migrate on these clones).
    4. Confirm with bd version; in server mode, bd doctor adds a migration-content-skew check (not available in embedded mode).

    A single clone with a remote follows the same gate (push with the old binary, then migrate and push). Full recipe and the single-clone variant: Upgrading bd — remote-backed databases and multiple clones. If clones have already forked, see the recovery playbook: docs/RECOVERY.md#pk-fork-refused.

Added

  • perf(schema): composite (status, updated_at) and standalone defer_until indexes on issues (migration 0052). Speeds up bd stale and bd ready on large rigs. First run after upgrade applies the migration — expect a one-time 10–25 s pause on rigs with >10K issues. Do not interrupt.

  • Per-migration content hash. schema_migrations now records the SHA-256 of each migration's file content alongside its version (content_hash), so two clones at the same MAX(version) but with divergent migration content become detectable (reporter fix No.2 for #4259). The column is added to fresh databases via the bootstrap schema and idempotently to existing databases at migrate time; already-applied rows keep a NULL hash. The column definition and the hashes are deterministic, so schema_migrations still merges cleanly across clones.

  • bd doctor migration-content-skew check. Using the recorded hashes, bd doctor now compares the local schema_migrations against the cached remote-tracking ref (no network fetch) and warns when this database and its remote applied different content for the same migration version — the silent schema fork from #4259, surfaced as a clear advisory instead of a cryptic merge failure. Read-only diagnostic (it does not gate push/pull); the comparison primitive (schema.ContentHashSkew) is reusable.

  • Remote-migrate prevention gate. bd now refuses to silently auto-apply pending schema migrations to an existing database that has a remote configured (in both server and embedded mode), and tells the operator to choose: migrate (as the single designated migrator, then bd dolt push) or adopt the already-migrated database from the remote (bd bootstrap). Migrating each clone independently forks the schema and breaks bd dolt pull (#4259). The gate is a no-op for fresh databases, databases already at the binary's version, databases with no remote, and read-only opens. The designated migrator proceeds with BD_ALLOW_REMOTE_MIGRATE=1. In server mode the gate also detects remotes persisted on disk in .dolt/config, so a freshly (auto-)started server — whose in-memory dolt_remotes table is not yet populated — cannot slip a remote-backed database past the gate.

  • Recovery guidance for Dolt primary-key merge refusals. When bd dolt pull (or push) fails with Dolt's hard refusal cannot merge because table X has different primary keys [in its common ancestor] — the un-mergeable schema fork left behind by independently-migrated clones straddling a PK-reshaping migration (#4259) — bd now recognizes the error class and prints the bootstrap-from-canonical-clone recovery recipe instead of just the raw error. Full playbook: docs/RECOVERY.md#pk-fork-refused.

Fixed

  • Cross-clone issue-delete vs child-row-insert merges now converge. The synced child tables (dependencies, labels, comments, events, snapshots) carry FOREIGN KEY ... ON DELETE CASCADE to issues (migrations 0041/0042), but Dolt merges row-wise and never re-executes cascades: if clone A deleted an issue while clone B added a dependency, label, or comment referencing it, bd dolt pull failed outright with a foreign-key constraint violation ("transaction rolled back"), recorded nothing in dolt_conflicts, and could never converge on retry. The pull auto-resolver (SQL and CLI routes) now repairs this class by applying the foreign key's own cascade semantics — deleting the merged-in rows whose issue reference dangles — and refuses, exactly as before, any constraint violation it does not recognize (different constraint type, table bd does not own, FK to another parent). The repair converges across clones: it deletes precisely the rows the cascade already removed on the deleting clone. (bd-6dnrw.4)

  • is_blocked recompute no longer misses auto-resolved merges. A pull whose merge bd auto-resolved (conflicts) or cascade-repaired (constraint violations) lands in the working set without advancing HEAD — the merge commit is created later — so the bd-6dnrw.3 post-pull recompute mistook it for "nothing merged" and skipped, leaving is_blocked stale exactly when the merge had needed intervention. The recompute now diffs to the working set and skips only when issues/dependencies are clean too. (bd-6dnrw.39)

  • is_blocked recomputed after bd dolt pull. The denormalized is_blocked column (migration 0046) was maintained only by local write paths, so a pull that merged another clone's writes silently left it stale — e.g. clone A closes blocker X while clone B adds an edge W→X, and the merged result carries W.is_blocked=1 with a closed blocker, hiding W from bd ready. Every pull path (server SQL and CLI routes, peer pulls, embedded mode) now recomputes the column for exactly the rows the merge changed, scoped via dolt_diff between the pre- and post-pull HEADs and expanded through the same affected-set logic the local write paths use. Oversized or schema-reshaping merges fall back to a full recompute; conflicted pulls skip it until the operator resolves. (bd-6dnrw.3, PR 4107 follow-up)

  • Stale is_blocked is now repairable — bd recompute-blocked. The post-pull recompute above is scoped to the merge diff and is skipped when a re-pull merges nothing (HEAD unchanged), so a recompute that failed after its merge committed — or a conflicted pull resolved by hand (which skips the recompute) — could leave is_blocked stale with no way to repair it: rerunning bd dolt pull merged nothing and recomputed nothing. The new bd recompute-blocked command runs a full, unconditional recompute over every issue and wisp and commits the result; it is idempotent and works in both embedded and server mode (unlike bd doctor, which is server-mode only). Server-mode bd doctor also gains a read-only Blocked State check that reports stale rows and a bd doctor --fix that runs the same repair. (bd-6dnrw.37)

  • Deterministic history-table primary keys (cross-clone merge-safety). Migration 0037 converted the legacy BIGINT primary keys of events, comments, issue_snapshots and compaction_snapshots by backfilling UUID() — a different random value on every clone, the same hazard class as the dependencies fork below. Two legacy clones that upgraded independently held identical history rows under different primary keys, so their merges duplicated every pre-upgrade event and comment (or refused outright). A one-time-per-clone upgrade backfill now rewrites those ids to deterministic values derived from each row's content (internal/storage/rowid), so independently-migrated clones converge to byte-identical history tables; exact-duplicate rows are kept distinct via per-duplicate ordinals. The pass is gated on a clone-local marker (ignored migration 0009), so steady-state opens never re-key — rows inserted after the pass are minted once and reach other clones by merge, so their ids are already consistent. (bd-6dnrw.2, the history-table sibling of #4259)

  • Deterministic dependency primary keys (cross-clone merge-safety). dependencies.id (and wisp_dependencies.id) were filled by DEFAULT (UUID()), a per-clone-random value. Two clones that created the same edge — or that applied migration 0043 independently — diverged on the primary key, so bd dolt pull failed unrecoverably (cannot merge because table dependencies has different primary keys in its common ancestor, or a uk_dep_* unique-key violation). id is now derived deterministically from the natural edge key (issue_id, target) at every insert site (internal/storage/depid); migration 0050 drops the random default and idempotently re-asserts the natural-identity unique keys; and an upgrade backfill rewrites existing rows. Independently-migrated clones now converge to byte-identical, merge-safe dependencies. And because the same edge now has the same primary key on every clone, bd dolt pull auto-resolves a same-edge dependency conflict that differs only in audit columns (created_at/created_by/metadata/thread_id) the same way it already does for the metadata table — so two machines that each run bd dep add X Y between syncs merge cleanly. A conflict where the dependency type differs is still surfaced for the operator. (#4259)

[1.0.5] - 2026-05-28

Upgrade Notes

  • JSONL auto-export and auto-staging are now opt-in. Repositories that relied on the previous implicit default should set it explicitly before or after upgrading:

    bd config set export.auto true
    bd config set export.git-add true
    

    Dolt is the primary datastore. .beads/issues.jsonl is now treated as an optional export for viewers such as bv, interchange, and issue-level migration. It is not the canonical git-tracked source of truth, not cross-machine sync, and not a full database backup. Use bd dolt push / bd dolt pull for sync and bd backup for restorable database backups. Existing repositories that already have export.auto: true or export.git-add: true configured keep that behavior.

Added

  • Schema-version guard: bd now hard-fails at database-open time when the database schema is ahead of the binary's embedded migrations. This protects against cryptic SQL errors (e.g., column X could not be found in any table in scope) that occur when an older binary opens a database migrated by a newer one. The error message names the exact version mismatch and provides a copy-paste rebuild command. Use --ignore-schema-skew (or env var BD_IGNORE_SCHEMA_SKEW=1) to bypass the guard when you know the forward migrations are safe for your workload. Fresh databases and normal forward-migration paths are unaffected.

  • Foreign keys across issue and wisp tables. Migrations 0040–0042 and the new ignored/0001–ignored/0004 add explicit FKs with ON DELETE CASCADE ON UPDATE CASCADE on dependencies, labels, comments, events, issue_snapshots, compaction_snapshots, child_counters, and the matching wisp_* tables. Deleting or renaming a parent row now cascades automatically — the manual cleanup loops in issueops/delete.go, dolt/wisps.go, dolt/ephemeral_routing.go, and cmd/bd/rename_prefix.go have been removed (net ~300 lines down). (#3952)

  • issueops.DeleteWispFromDependenciesInTx / UpdateWispIDInDependenciesInTx. Because Dolt forbids foreign keys from tracked tables (dependencies) to dolt_ignore'd tables (wisps), wisp deletion and rename now invoke these helpers explicitly to keep dependencies.depends_on_wisp_id consistent. The standard store APIs (DeleteIssue, UpdateIssueID, deleteWispBatch, etc.) wire them up automatically; only call them directly if you bypass those entry points. (#3952)

  • Forward schema-skew guard. bd now hard-fails when it opens a database that has been migrated to a newer schema version than the binary understands, instead of operating blindly on forward-migrated data. (#4152)

  • dolt.mode config key. New dolt.mode (server | embedded) configuration key with validation; bd init now warns on ambiguous configs and hard-fails when dolt.host/dolt.port are set without server mode.

  • bd list --skip-labels. New toggle to skip label hydration for faster listings; honored in both text and JSON output.

  • bd show count-only JSON details. bd show --json can return count-only relationship details with opt-in streamed payloads, avoiding large hydrations when only counts are needed.

  • Server-side custom type validation. Storage now honors .beads/config.yaml types.custom during server-side validation. (#4024, #4026)

  • bd init contributor routing. On fork detection, bd init auto-configures contributor routing. (#4028)

Changed

  • Large content columns widened to LONGTEXT. Issue content and event/wisp-event value columns are migrated from TEXT to LONGTEXT, lifting the previous ~64 KB per-field ceiling. (#4027, #3682)

  • dependencies.depends_on_id is now a STORED generated column. The polymorphic target has been split into three typed columns: depends_on_issue_id, depends_on_wisp_id, depends_on_external. depends_on_id remains as COALESCE(...) STORED for read paths; writes to depends_on_id will fail — code that inserts dependencies must populate exactly one typed column (enforced by a new ck_dep_one_target CHECK). Same split mirrored to wisp_dependencies with a corresponding ck_wisp_dep_one_target. Migrations 0041 (tracked) and ignored/0003 (wisps) perform the column split, copy existing rows by classifying their targets against issues / wisps, and add the new typed-target indexes. (#3952)

  • Most existing dependency-table FKs now use ON UPDATE CASCADE. Migration 0042 rebuilds fk_dep_issue, fk_labels_issue, fk_comments_issue, fk_events_issue, fk_snapshots_issue, and fk_comp_snap_issue to cascade on both delete and update. Same treatment for the wisp-side FKs in ignored/0003. Prefix rename and ID-update paths rely on this cascade instead of touching aux tables manually. (#3952)

  • Migrations 0041 and 0042 are intentionally irreversible. The matching .down.sql files are documented no-ops because rebuilding the polymorphic column from typed columns would require schema-aware backfill that the storage layer no longer performs. Restore from a prior dolt commit if rollback is required. (#3952)

  • BEADS_MAX_ROWS / --max-rows defensive row cap — ops can now set BEADS_MAX_ROWS=N (or pass bd list --max-rows N) to bound SearchIssues result sets. Default is disabled; on overage, bd exits with code 2 and writes a two-line error to stderr (stdout stays empty so jq pipelines don't get half-rendered JSON). The flag is wired on bd list, bd ready, bd dep tree, bd find-duplicates, and bd graph. The env var is also honored by bd doctor, bd lint, bd doctor-conventions, and bd doctor-pollution. bd cleanup, bd gc, bd export, bd export --auto, bd migrate-issues, and bd jira explicitly opt out so a misconfigured env can't abort a sweep partway. (be-x42v)

Fixed

  • Symlinked CLAUDE.md repair guidance - troubleshooting docs now include a tested git update-index --cacheinfo recipe for repositories that already contain a corrupted mode 120000 CLAUDE.md entry with Markdown content. The setup bug was fixed in #4192; this note is for repairing affected clones.
  • Codex hook metadata isolation - Codex hook metadata now lives under .codex-plugin/hooks/, preventing Claude from loading Codex-only hooks from the shared plugin root while keeping Codex setup pointed at bd codex-hook. (#3924)
  • Hook and prime wait bounds - generated git hooks now fall back from timeout to gtimeout to Perl alarm, and bd prime memory loading is bounded by BEADS_PRIME_TIMEOUT (default 10s). Non-positive BEADS_PRIME_TIMEOUT values fall back to the default instead of silently disabling the safety bound. Internal beads-repo git operations also run with hooks suppressed via core.hooksPath= so automated .beads/ commits and pushes cannot recurse through user hook managers. GitCmdCWD() remains for user-working-repo status/ref operations and does not perform beads commits. (#4172)
  • JSONL auto-export shrink guard - when opt-in export.auto is enabled, auto-export now refuses to overwrite an existing .beads/issues.jsonl that contains records outside the auto-export scope (memories, infrastructure beads, templates, ephemeral wisps, or unknown record types). This prevents a viewer/interchange refresh from silently replacing a richer JSONL file with the filtered auto-export subset. (#4069)
  • bd quick now treats label insertion as part of issue creation: label failures abort the create instead of being silently dropped.
  • bd dolt status reports externally-managed local servers truthfully - when a rig is configured as dolt_mode: server pointing at a local host but dolt.auto-start: false (so an orchestrator or systemd owns the sql-server lifecycle), bd dolt status previously said not running because no PID file existed. It now SQL-probes the configured endpoint, matching the path already used for non-local hosts, and reports running (external) with host/port/database/version when the server answers. JSON output shape change: on affected rigs, bd dolt status --json now emits {"running": true, "mode": "external", ...} instead of {"running": false, "pid": 0, ...}. Automation that parsed the old running:false as a "needs restart" sentinel should switch to checking running directly. (be-0eyj, #3550)
  • Migrations run on a connection with no read timeout, so large or slow schema migrations no longer fail under the default query timeout. (#4154)
  • Duplicate migration version numbers now hard-fail instead of silently under-applying a release's migrations. (#4197)
  • bd close supports per-id reasons when closing multiple issues at once (#4194), and re-closing an already-closed issue is now an idempotent success instead of an error. (#4025)
  • bd create --defer <future-date> creates a deferred issue (status deferred) rather than an open one. (#4071)
  • bd create commits labels atomically during initial issue creation; label-insertion failures abort the create instead of being silently dropped. (#4149)
  • JSONL import preserves comment IDs and skips stale issue snapshots. (#4103, #4204)
  • bd init Dolt wiring hardening - honor local-only remote wiring (#4227), apply config server mode before validation, platform-accurate config-path warnings, and tighter Dolt marker handling (#4083).
  • MCP validate and detect-pollution route to bd doctor. (#4037)
  • bd dolt detects managed handoff port conflicts. (#4217)
  • bd list keeps relates-to out of dependency tree mode and honors --skip-labels in JSON output. (#3936)
  • Audit newID entropy widened to reduce ID-collision risk under high create volume. (#4046)
  • Performance: bd count uses a SQL COUNT(*) aggregate instead of fetching and counting rows; bd ready narrows deferred-parent child filtering; bd get queries primary issues before falling back to wisps; and bd search gains skip-wisps merge and wisp-count caching.

[1.0.4] - 2026-05-07

Added

  • bd init --reinit-local / --discard-remote — named-intent flags for local re-initialization and explicit remote-history override. Replaces the overloaded --force. See bd help init-safety and docs/recovery/init-safety.md.
  • bd init-safety — documents the init flag surface + destroy-token format. Referenced by every init refusal message.
  • Stable exit codes for init refusals — 10 remote divergence, 11 local exists, 12 destroy-token missing. Grep-safe for CI.
  • ADR 0002 — bd init safety invariants — encodes the single-source identity rule, scope-bound --force/--reinit-local, the CheckRemoteSafety chokepoint, the error-text-no-echo rule, and the race-safety invariant.
  • docs/RECOVERY.md — playbooks for each named init refusal.
  • CODEOWNERS — cmd/bd/init*.go routes review to maintainers with an ADR-linked acknowledgment requirement.
  • bd -C <path> — run bd from another directory without changing the caller's shell cwd. Useful for hooks, agents, and scripts that coordinate multiple workspaces.
  • bd close --reason-file — reads close reasons from a file or stdin, matching existing body-file workflows.
  • Linear sync throughput and correctness improvements — batch create/update, idempotency markers, retry-after handling, OAuth client-credentials support, and workspace-level sync locking.
  • Shared Beads plugin package — Claude and Codex plugin metadata now ship together from the repository.

Changed

  • bd init --force semantics narrowed to local-only — --force (or --reinit-local) bypasses only the LOCAL data-safety guard. It does NOT authorize silent divergence of remote history. When origin has refs/dolt/data, bd init --force now refuses with exit code 10 unless --discard-remote is also passed. Fixes the long-standing footgun where bd init --force in a repo with remote Dolt history silently set up an orphan branch that failed to push.

  • Init refusal messages follow What/Why/Next structure — runtime error text no longer echoes copy-pasteable destructive invocations. Token values and exact override commands live in bd help init-safety and docs/RECOVERY.md only. Closes the failure class where an AI agent destroyed 247 issues by pattern-matching on the tool's own error output (58f5989bf).

  • beads.OpenBestAvailable signature changed (breaking) — returns (Storage, error) instead of (Storage, Unlocker, error). The per-open flock and Unlocker return value were removed; the embedded Dolt engine handles its own concurrency internally. External SDK consumers that call OpenBestAvailable must drop the second return value. (PR #3614)

  • Embedded-mode flock removed — the process-lifetime exclusive flock on .beads/embeddeddolt/ has been removed. Concurrent bd processes now open the embedded engine independently. If the GH#2571 nil-deref stack (NewConnector → DoltDB.SetCrashOnFatalError → CollectDBs) resurfaces under concurrent access, it should be filed and fixed in dolthub/driver, not reintroduced as a beads-side flock. (PR #3614)

  • Auto-export is now opt-in by default — repositories without explicit export.auto / export.git-add settings leave JSONL refresh and staging disabled unless a viewer or JSONL integration explicitly enables them. This keeps .beads/issues.jsonl as an optional export surface, not the default mutation path. (GH#4062) Thanks Kevin Glynn for surfacing the JSONL source-of-truth mismatch through the export cleanup work that led to this change.

  • Release workflow uses the checked-in beads-release formula — the old release shell path now delegates to the formula-backed release workflow with CI gates.

Deprecated

  • bd init --force — deprecated alias for --reinit-local. Continues to work for ≥2 releases, emits a DeprecationWarning, routes internally to --reinit-local. Use --reinit-local going forward. CI and scripts are unaffected for the duration of the deprecation window.

Fixed

  • bd close <id> in contributor auto-routing mode — bd close now resolves IDs through the same routing fallback as bd show and bd update. Previously bd close errored with no issue found matching <id> in workspaces with routing.mode=auto, routing.contributor=<path>, and beads.role=contributor — even for IDs that bd create had just returned, because the issue lived in the routed planning store while bd close only checked the local primary. Bulk close (bd close <id1> <id2> ...) is also fixed by sharing one routed-store handle across the close batch. Fixes #3608.
  • Dolt-in-git hook recursion — internal pushes of refs/dolt/data skip git hooks, preventing sync loops. Fixes #3724.
  • Hook setup and migration safety — generated hooks no longer drop sibling commands, stale Claude hooks are warned about, and post-pull/checkout auto-import keeps committed .beads/issues.jsonl state fresh.
  • Dependency display and routing — dependency trees show blocked state only for genuine blockers, tolerate unresolved IDs in batch output, and include dependency type information.
  • Export filtering — closed memories and ephemeral wisps are excluded from default exports, and NoHistory beads survive explicit filters.
  • Graph/list regressions — graph HTML emits empty arrays instead of null, merges disconnected components for --all --html, and list output avoids duplicate issues with multiple blockers.
  • Configuration documentation and validation drift — stale TOML/BEADS_* references were corrected to YAML/BD_*, and secret keys are refused in git-tracked config.
  • Release preflight and packaging — release formula adoption, script selection, pure-Go test helpers, Windows install cleanup, and generated CLI reference paths were hardened for the v1.0.4 release attempt.

[1.0.3] - 2026-04-24

Fixed

  • go install ...@latest restored — released a tag whose go.mod no longer contains local replace directives, fixing install failures from v1.0.1 and v1.0.2. Fixes #3338 and #3312.
  • npm macOS archive layout — macOS release archives are flat, matching the path expected by the npm postinstall script. Fixes #3388.
  • Windows install cleanup — make install on Windows removes stale extensionless bd files before installing bd.exe, avoiding PATH shadowing by older installs. Fixes #3399.
  • Windows test portability — fixed tests that hardcoded POSIX path or permission behavior. Fixes #3404.
  • Embedded pre-commit hook cwd — pre-commit export no longer prints "no beads database found" for embedded-mode repositories. Fixes #3454.

[1.0.2] - 2026-04-15

Fixed

  • npm publish: provenance validation — npm-package/package.json repository.url, bugs.url, and homepage updated from steveyegge/beads to gastownhall/beads so that npm's sigstore provenance validator accepts the published artifact. The repo move in v1.0.0 left these URLs stale, which caused npm publishes for v1.0.0 and v1.0.1 to be rejected after provenance check (E422).

[1.0.1] - 2026-04-15

Added

  • bd batch — Atomic multi-operation transactions across create/update/close/dep. All ops commit together or none do. (PR #3165)
  • bd config drift / bd config apply — Detect and reconcile drift between yaml, git, and database config surfaces. (PR #3086)
  • bd config show — Unified provenance view showing where each effective config value comes from. (PR #3084, bd-934)
  • started_at on issues — Timestamp recording when an issue first entered in_progress. (PR #3206, GH#2796)
  • beads_global shared-server database — Dedicated shared-server database for cross-project state. (fa87dd2b)
  • Pool metrics telemetry — Shared-server connection pool metrics for diagnosing pool exhaustion. (PR #3172, GH#3140)
  • OpenBestAvailable public API — Library consumers can open the best available backend without hard-coding selection. (PRs #3143, #3149)
  • Selective sync: --issues flag and push/pull subcommands — Tracker sync accepts explicit issue ID lists; --parent ported across trackers. (PR #2975)
  • --label / --label-any for bd orphans — Filter orphan detection by label. (PR #3026)
  • Cross-version smoke tests — CI gate for upgrade compatibility across prior releases. (GH#2968)
  • Migration test harness — Cross-era upgrade fidelity tests. (PR #3067)
  • Docusaurus versioning — Site versioned per release; llms-full.txt aligned with snapshots. (PR #3033)
  • Azure Blob Storage remotes — az:// recognized as a valid Dolt remote URL scheme. (PR #3101)
  • Side-effect hints for bd config set/unset — Shows what else changes when a config value is modified. (PR #3089)
  • Dynamic config-list discovery — bd config list auto-discovers yaml keys and env vars. (PR #3088)
  • BEADS_DOLT_READY_TIMEOUT — Override the 10s waitForReady timeout during bd init --shared-server for slower hardware. (PR #3188, GH#3142)

Changed

  • Auto-export enabled by default — export.auto defaults to true and the default export.path is now issues.jsonl (previously export.jsonl), with export.git-add on by default. bd init prompts interactively (default: keep enabled); --non-interactive keeps it enabled. All export.* keys are listed in bd config --help. (PR #3204, GH#2973)
  • gms_pure_go by default — Test and install helpers default to the pure-Go build tag, dropping ICU linkage from CI. (PRs #3240, #3259)
  • ICU runtime dependency removed from release binaries — Release builds no longer link ICU, fixing portability issues. (PR #3066)
  • Dolt connection pool lifetime configurable — Longer default, configurable via env. (PR #3163)
  • Dolt connection TLS explicitly disabled — When TLS is not configured, connections now explicitly opt out rather than depending on driver defaults. (PR #3107)
  • Dolt log noise reduced — NewConnection spam silenced; dolt-server.log rotates at startup to cap size. (PRs #3160, #3161)

Fixed

  • Schema migration conflict — Removed a conflicting column on the schema migration table; migration issue patched with test coverage. (commits f079786b, 65dbdbf5, 9aa9b82f)
  • Wisp tables missing after bootstrap clone — bd bootstrap now ensures wisp tables exist after cloning. (commit 31d51232)
  • CI test binary timeout on macOS — bd test binary memoized to avoid 10-minute timeout. (PR #3273)
  • Local-only state moved to Dolt-ignored tables — Prevents local state from polluting shared history. (commit 50f715b1)
  • Worktree awareness across commands — Hooks path, fingerprint, doctor checks, config validate, preflight, reset, bootstrap path synthesis, rename-prefix, countExistingIssues, and formula search paths all resolve correctly when run inside a git worktree. (PRs #3169, #3123, #3235, plus numerous smaller fixes)
  • bd bootstrap hardening — Commits issue_prefix to the config table; writes metadata.json and config.yaml after sync clone; uses parent workspace db name when local .beads missing; rejects empty --database; auto-start allowed for migration tests. (PRs #3203, #3247, #3083, #3076)
  • Remote URL validation hardened at config parse time — Security hardening for tracker and backup remotes. (PR #3210)
  • bd list truncation hint in all output modes — Truncation now indicated in JSON/compact/long formats. (PR #3243, GH#3212)
  • bd list --watch hierarchy consistency — Parent/child ordering stable across watch refreshes. (PR #3236)
  • bd update --defer sets status=deferred — Deferring an issue now transitions status correctly. (PR #3241, GH#3233)
  • bd mol bond transitive cycle detection — Cycles through transitive dependencies now caught. (PR #3111, GH#2719)
  • bd dep add/remove allows cross-prefix targets — No longer rejects dependencies that cross prefix boundaries. (commit 7b02edda)
  • bd dolt pull nil-pointer panic in embedded mode — Embedded backend no longer panics on pull. (PR #3148)
  • Diverged-history guidance on auto-push failure — Actionable recovery steps printed instead of raw error. (PR #3138)
  • bd config error message hints — Suggest bd config set (not bd config) in error messages. (PR #3141)
  • JSONL export in pre-commit hook — Atomic code+issues commits via pre-commit hook. (PR #3121)
  • Auto-export / auto-backup errors surfaced to stderr — Silent failures now visible. (PR #3122)
  • go install on Windows — Fixed ICU header dependency that broke go install on Windows. (PR #3112, GH#3013)
  • Stepping-stone migration paths removed — Unsupported intermediate migration versions cleaned up. (PR #3110)
  • SQLite-era JSONL migration preserves dependencies and labels — Migration from pre-Dolt exports no longer drops graph edges or labels. (PR #3082, GH#3079)
  • MCP workspace discovery detects Dolt-backed projects — MCP server no longer misses Dolt-native repos. (PR #3207, GH#2997)
  • npm postinstall closes download streams — Resource leak fixed. (PR #3228)
  • Husky hooks sanitized when copied — Husky-managed hooks no longer corrupt beads-managed hook directory. (PR #3208, GH#3132)
  • GitLab issue link dependencies wired into sync pull — Dependencies via GitLab links now import. (PR #3202, GH#2645)
  • .beads/ FS_NOCOW_FL on btrfs — Prevents btrfs kworker thrashing on Dolt files. (PR #3162)
  • Auto-commit config table writes — bd remember, bd forget, and bd config commands now commit their writes. (PR #3052, bd-g8p)
  • Shared-server CLI dir resolved from shared root — Corrects path resolution for shared-server layouts. (PR #3223)
  • .claude/ gitignore narrowed — Blanket ignore replaced with specific patterns so project .claude/ content is tracked. (PR #3190, GH#3182)
  • Release archives extracting into subdirectory — Install handles nested archive layouts. (PR #3167)
  • bd import help text — Removed references to nonexistent -i flag. (PR #3200)

Docs

  • ADR-0001: multi-remote approach decision (PR #3209)
  • ICU regex policy / build dependency guidance (GH-3126)
  • Contributor protection policy for AI agents (PR #3151)
  • Cross-era migration instructions (PR #3081)
  • Unified quick-start (site + stub) (PR #3032)
  • Hosted docs link (PR #3011)

[1.0.0] - 2026-04-02

1.0 — Stable Release

Beads 1.0 marks the transition from rapid iteration to production stability. The Dolt migration that began in v0.55.0 is complete: embedded Dolt is the default on all platforms, server lifecycle management is no longer required, and the critical reliability issues that affected the v0.55–v0.63 series have been resolved.

Added

  • Embedded Dolt default on macOS — Native CGO builds on macOS enable embedded Dolt by default, eliminating the external Dolt server entirely. (PR #2971)
  • Custom status/type normalized tables — custom_statuses(name, category) and custom_types(name) tables replace comma-separated config strings. Views use table-backed subqueries instead of dynamic IN clauses. Migration is automatic and idempotent. (PR #2961)
  • bd rules audit and bd rules compact — Scan .claude/rules/*.md for contradictions and near-duplicates; merge rule groups with deduplication. (PR #2810)
  • bd config set-many — Batch config operations across yaml, git, and database keys in a single command. (PR #2943)
  • Credentials file support — ~/.config/beads/credentials for Dolt server passwords with Unix permission checks, env var override, and proper DSN escaping. (PR #2854)
  • Spike, story, and milestone issue types — First-class support with constants, validation, and normalization aliases. (PR #2923)
  • GitLab sync: epic→milestone, task hierarchy, type filtering — Epics sync as GitLab milestones, tasks create work items via GraphQL with parent hierarchy, internal types excluded from push by default. (PR #2889)
  • ADO shorthand refs — ado:NNNNN recognized in IsExternalRef and ExtractIdentifier, preventing duplicates on sync. (PR #2966)
  • ADO push filters — --types, --states, and --no-create flags now apply to push operations. (PR #2944)
  • Public format package — format.PrettyIssue, CompactIssue, LongIssue for external tooling. (PR #2871)
  • SlotSet/SlotGet/SlotClear — Per-issue metadata slots in the Storage interface for delegation tracking and hook state. (PR #2870)
  • WispFilter export — beads.WispFilter type alias exported from root package. (PR #2868)
  • --non-interactive / --yes for bootstrap — Skip confirmation prompts in CI/automation. (PR #2942)
  • Batch dependency list — GetDependenciesWithMetadata supports multiple issue IDs. (PR #2875)
  • bd ready --explain — Dependency-aware reasoning for why issues are or aren't ready. (PR #2814)
  • Graph integrity enforcement — Extended cycle detection and bd graph check command. (PR #2814)
  • Tracker hardening — Retry logic with jitter for all tracker HTTP clients, pagination guards for Linear/Jira, response size limits, terminal-safe content sanitization. (PRs #2814)
  • Release stability gate — Upgrade safety checks and breaking-change communication. (GH#2951)
  • Hook firing in storage layer — HookFiringStore decorator moves hook logic out of CLI layer. (PR #2891)

Changed

  • bd setup claude defaults to project-local — Settings written to .claude/settings.json (shared/team) by default; --global for user-level. Auto-runs during bd init. (PR #2972)
  • strconv.ParseBool for env vars — IsAutoStartDisabled uses standard boolean parsing with whitespace trimming. (PRs #2955, #2956)
  • Comma-separated --status filter — bd list --status open,in_progress now works. (GH#2846)
  • Custom types from YAML — ResolveCustomTypesInTx honors config.yaml custom types in create path. (GH#2793)
  • lipgloss v2, backoff v5, goldmark 1.8.2 — Dependency updates.

Fixed

  • Concurrent embedded Dolt panic — Exclusive flock acquired on embedded store open, held for store lifetime. Second concurrent opener gets a clear error instead of a nil-pointer panic. (GH#2571)
  • Spaces in repo path — Embedded Dolt no longer URL-encodes local filesystem paths, fixing %20 lookup failures on macOS. (GH#2920)
  • Schema version bump — currentSchemaVersion bumped to 11 for custom_statuses/custom_types migration. Prevents "table not found" errors on upgrade. (GH#2974)
  • Dolt init failure — Embedded-by-default eliminates server port 0 / connection refused issues for new users. (GH#2656)
  • Bootstrap for missing database — bd bootstrap probes server for actual database existence before declaring no-op. Recovery guidance updated across all error surfaces. (PR #2940)
  • Diverged history recovery — Dolt push/pull detect diverged history and print actionable recovery guidance. (PR #2941)
  • No-db command routing — bd context, bd dolt show, etc. honor --db flag, BEADS_DB env, and external dolt_data_dir layouts. (PR #2844)
  • Auto-backup address conflict — Detects when another backup remote owns the URL and syncs through it. (PR #2927)
  • Shared server project_id — bd init --database on a shared server adopts existing _project_id instead of generating a new one. (PR #2925)
  • Init leaves beads.role unset — Fixed edge cases where role was not persisted during init. (GH#2950)
  • --shared-server mode detection — Correctly treated as server mode for DoltMode resolution. (PR #2947)
  • GitLab work_items URL dedup — Pull dedup checks both issues and wisps tables; URL pattern matches /work_items/ paths. (PR #2889)
  • GitLab work item type ID — Dynamic GraphQL query with per-session cache instead of hardcoded gid://.... (Follow-up to PR #2889)
  • Stale noms LOCK files — Cleaned after bootstrap clone. (PR #2913)
  • ADO WIQL date precision — RFC3339 timezone formatting in WIQL queries. (PR #2911)
  • Blocked icon in bd list — Dependency-blocked issues now show blocked indicator. (GH#2858)
  • Installer ICU/CGO fallback — Correct handling on systems without ICU libraries. (PR #2965)

Community

Contributors: harry-miller-trimble, coffeegoddd, quad341, Bella-Giraffety, maphew, rjc123, Fredkr, sammywachtel, anupamchugh, ktoulgaridis, outdoorsea, Geargrindadmin.

[0.63.3] - 2026-03-30

Added

  • Embedded mode by default — New installations now use embedded Dolt as the default storage backend, eliminating the need to run a separate Dolt server process. Existing server-mode installations are unaffected. (PR #2842)
  • bd create --graph — Batch issue creation from a YAML/JSON dependency graph, replacing the standalone bd graph-apply command. (PR #2894)
  • Multi-project/team sync — Linear, Jira, and ADO integrations now support syncing multiple projects and teams in a single repo. (PR #2868)
  • Notion sync — Reimplemented Notion integration on REST data sources with full page property mapping. (PR #2843)
  • GitLab sync — Optional group-level GitLab issue sync. (PR #2801)
  • Linear --parent flag — bd linear sync --parent limits sync push to a specific ticket tree. (PR #2853)
  • Remote URL support — Multi-repo hydration now supports remote URLs. (PR #2827)
  • Top-level command aliases — bd comment, bd assign, bd tag, bd link, bd priority as convenient shortcuts. (GH#2611)
  • Storage interface expansion — ListWisps, MergeSlot, ReopenIssue, UpdateIssueType, replication, and VC interfaces now exposed in the public API. (PRs #2782-#2788)
  • SearchIssues dependency hydration — Bulk dependency population for search results. (PR #2773)
  • --agents-file flag — bd init --agents-file customizes the agent instructions filename. (PR #2761)
  • Multi-process concurrency tests — Server lifecycle and concurrent access integration tests. (PR #2767)

Changed

  • Backup system simplified — Removed git-based backup/restore (backup export-git, backup fetch-git). Backup and restore now operate exclusively through Dolt's native replication. (PR #2842)
  • dolthub/driver v1.84.0 — Updated from v1.83.8, picking up serialization retry support.
  • Dolt write retry — writeRetryTx automatically retries on Dolt serialization errors. (PR #2855)
  • CI overhaul — Test sharding, precompiled binaries, shared Go module cache for faster CI runs.
  • Gas Town terminology removed — Remaining Gas Town naming removed from beads orchestrator detection. (PR #2762)
  • TypeScript v6, backoff v5, goldmark 1.7.17 — Dependency updates.

Fixed

  • Init git cache reset — bd init now resets git caches after auto-init, preventing stale .beads/dolt/ references in embedded mode. (PR #2907)
  • Doctor embedded mode — Improved embedded mode detection and push error guidance. (PR #2910)
  • Create description warning — Removed spurious description warning; test-issue warning now gated on DB size. (PR #2908)
  • External contributor PRs — block-gh-watch hook now allows external contributors to create PRs. (PR #2907)
  • CWD .beads/ priority — bd now checks CWD for .beads/ before git-worktree resolution, fixing worktree-based discovery in nested projects.
  • Epic closure eligibility — Wisp children now correctly included in epic closure checks. (PR #2843)
  • Prefix routing — bd edit, bd reopen, and bd delete now correctly route with prefixes from town root. (PRs #2780, #2847)
  • ADO/Jira custom type_map — Custom type maps no longer silently ignored. ADO dependency links now imported correctly. (PR #2771)
  • Init safety guard — bd init --force no longer bypasses close --quiet check. (PR #2799)
  • Circuit breaker TTL — Stale circuit breaker files cleaned during init; files moved to subdirectory. (PR #2799)
  • bd show --watch — Replaced fsnotify with polling to fix cross-platform compatibility. (PR #2755)
  • Config unset routing — config unset now routes to correct store; backup keys shown in list. (GH#2727)
  • Doctor false errors — Server health checks skipped when no Dolt server expected; bare worktree fallback fixed. (GH#2722)
  • UTC timestamps — ADO/Jira defer_until comparisons use UTC_TIMESTAMP(). (PR #2819)
  • Cloud storage routing — Push/pull routed through CLI when cloud storage env vars are set. (GH#2820)
  • Embedded schema — Removed HOP columns (crystallizes, quality_score) from embedded schema. (PR #2837)
  • JSON path quoting — Dotted metadata keys now properly quoted in JSON path expressions. (PR #2865)
  • Hook fire on update — bd update --status closed now fires on_close hook; bd close fires on_update. (PR #2754)

Embedded Dolt

  • 46 commits from the Dolt team (coffeegoddd) completing embedded Dolt support: bootstrap, compact, migrate, sync, backup, restore, maintenance, setup, config, and advanced commands. Includes serialization retry, driver bump to v1.84.0, CI test sharding, and RunInTransaction refactor for shared issueops operations.

Community

Contributors: coffeegoddd (Dustin Brown), matt wilkie (maphew), harry-miller-trimble, quad341, ktoulgaridis, eitsupi, renezander030, REMvisual, aphexcx, anupamchugh, MovGP0, Bella-Giraffety, julianknutsen, osamu2001, toumorokoshi, fernanja, sammywachtel, outdoorsea.

[0.62.0] - 2026-03-21

Added

  • Custom status categories — bd config set status.custom "in_review:active,qa_testing:wip,archived:done,on_hold:frozen" assigns categories to custom statuses that control behavior in bd ready (active statuses included) and bd list (done/frozen statuses excluded by default). Backward compatible — old flat format "foo,bar" still works with no behavior change. (PR #2665)
  • bd statuses command — Lists all built-in and custom statuses with their icons, categories, and descriptions. Supports --json for programmatic use.
  • bd note command — Shorthand for appending notes to issues without the bd update --note ceremony.
  • Audit logging — Status, assignee, and priority changes are logged to .beads/interactions.jsonl with close reasons. Survives Dolt GC flatten, so you can recover context after aggressive cleanup.
  • --exclude-type flag — bd ready --exclude-type wisp and bd list --exclude-type wisp filter issues by type.
  • --agents-profile flag — bd init --agents-profile minimal controls the default AGENTS.md profile (full/minimal).
  • Memories in export/import — bd export/bd import now round-trip agent memories.
  • Azure DevOps integration — New bd ado commands for syncing work items with Azure DevOps (bd ado sync, bd ado status, bd ado projects).
  • MCP ready tool filter — issue_type filter exposed on the MCP ready tool for agent use.
  • Quality/lifecycle commands in prime — bd prime, templates, and bd doctor now surface quality and lifecycle commands.

Changed

  • BEADS_ACTOR replaces BD_ACTOR — BEADS_ACTOR is now the primary environment variable for agent identity; BD_ACTOR remains as a deprecated fallback.
  • ServerMode enum — Dolt server ownership inference consolidated into a clean ServerMode enum (Auto, External, Shared, Embedded), replacing ad-hoc string checks.
  • dolthub/driver v1.83.8 — Updated from v0.2.1, picking up Dolt engine improvements and bug fixes.
  • anthropic-sdk-go v1.27.1 — Updated from v1.27.0.
  • Internal cleanup — Removed Gas Town-specific concepts from beads types and code: GUPP references, polecat/crew/overseer terminology, HOP fields, agent-as-bead subsystem, patrol molecule references, role templates, hardcoded ~/gt/ paths. Beads is now fully standalone.

Fixed

  • Windows Dolt server lifecycle — ErrServerNotRunning sentinel error prevents false "failed to stop" warnings; stale PID/port files cleaned up even when server is already dead. (GH#2670)
  • Hook preservation — bd init now preserves ALL pre-existing hooks (not just beads-managed ones) when overriding core.hooksPath. (PR #2738)
  • Shim timeout — Hook shim timeout increased from 30s to 300s, preventing premature termination of chained pre-commit pipelines (linters, formatters, type-checkers). Configurable via BEADS_HOOK_TIMEOUT. (GH#2732)
  • GetNextChildIDTx query — Replaced broken CAST(SUBSTRING_INDEX(...) AS UNSIGNED) with Go-side ParseHierarchicalID, fixing child ID reconciliation for non-numeric IDs after JSONL imports. (GH#2721)
  • Credential key in .gitignore — .beads-credential-key added to project-root .gitignore template as defense-in-depth for the AES-256 encryption key. (GH#2695)
  • SQL stored procedures for flatten/compact — bd flatten and bd compact now use SQL stored procedures in server mode, fixing failures on shared Dolt servers.
  • --format json collision — Resolved flag collision between rootCmd persistent --format flag and list/dep-tree local --json flags.
  • Concurrent schema init — initSchemaOnDB serialized with GET_LOCK to prevent Dolt journal corruption in multi-agent environments. (GH#2672)
  • External Dolt server safety — KillStaleServers respects IsAutoStartDisabled and server ownership, preventing accidental kills of externally-managed servers. (GH#2641)
  • Doctor infinite restart loop — Doctor no longer triggers infinite Dolt server restart cycles. (GH#2636)
  • Schema version 8 — Dolt schema bumped to version 8 with proper commit of version update. (PR #2635)
  • Repo-local servers — Auto-started Dolt servers kept alive across sequential bd commands within the same repo.
  • Store config preservation — Reopening stores by path no longer drops repo-specific config. (PR #2628)
  • Validation on close — --validate now checks --acceptance field; validation.on-close config controls enforcement. (PR #2654)
  • .beads/.env load order — Environment file loaded before noDbCommands early return, fixing env-dependent non-DB commands. (GH#2677)
  • Update description safety — Empty stdin/file no longer silently erases descriptions; requires explicit opt-in.
  • Bootstrap JSONL imports — Correctly imports from git-tracked .beads/issues.jsonl.
  • Doctor fail-closed — Server-mode integrity recovery now fails closed instead of silently passing.

Embedded Dolt

  • 73 commits from the Dolt team (coffeegoddd) advancing embedded Dolt support: create, list, update, close, show, delete, search, query, label, gate, promote, move, merge-slot, quick, diff, count, find-duplicates, graph, and views/reports commands now work in embedded mode. Includes shared issueops package extraction for transaction-based operations used by both server and embedded backends.

Community

Contributors: coffeegoddd (Dustin Brown), matt wilkie (maphew), harry-miller-trimble, gzur, Algorune, sfncore, angelamayxie, paf0186, Patrick Farrell, Tim Visher.

[0.61.0] - 2026-03-15

Added

  • --no-history flag — bd create --no-history stores beads in the wisps table (skipping Dolt commits) without making them GC-eligible, fixing agent identity beads being incorrectly garbage collected. bd update --no-history/--history toggles the flag. (GH#2619)
  • --claim-next flag — bd close --claim-next auto-claims the next highest priority available issue after closing
  • --skills and --context flags — bd create --skills/--context add structured sections to issue descriptions
  • --exclude-type for wisp GC — bd mol wisp gc --exclude-type agent,rig protects specific types from garbage collection
  • --type filter for wisp list — bd mol wisp list --type bug filters by issue type
  • Auto-detect beads database on git origin — bd bootstrap probes origin for refs/dolt/data and auto-clones the database; bd init warns when existing data is detected (GH#2580)
  • Install checksum verification — install scripts verify SHA256 checksums before extraction (GH#1857)
  • Jira pull_jql config — jira.pull_jql config key or JIRA_PULL_JQL env var for custom JQL filters during issue fetch
  • UUID primary keys — events, comments, and snapshot tables migrated from AUTO_INCREMENT to UUID() primary keys for federation-safe multi-clone operation

Changed

  • Prime SSOT — bd prime is now the single source of truth for agent instructions. Agent files use versioned hash-based markers with profile support (full/minimal) for staleness detection and automatic upgrades. Symlink safety and profile precedence (full preserved over minimal) (GH#2139)
  • Formula extends merges by ID — child steps with matching IDs override parent steps in-place instead of appending duplicates
  • Circuit breaker keyed on host:port — prevents cross-host blocking when servers share a port
  • Backup scoped by prefix — backup export and restore filter by current project prefix, preventing cross-project data leakage on shared servers

Fixed

  • Embedded Dolt CI — clean stale noms LOCK files before opening embedded store (fixes TestEmbeddedInitConcurrent panic), update stale from_jsonl test, gofmt import.go
  • Circuit breaker port-0 poisoning — returns nil breaker for unresolved port 0 during standalone auto-start
  • KillStaleServers scoped to repo — only kills orphan Dolt processes belonging to the current repo's data directory
  • Port resolution chain — applyConfigDefaults consults port file > config.yaml > metadata.json instead of falling through to default 3307
  • Doctor fingerprint checks — bd doctor /path/to/repo computes fingerprints from the target path, not cwd
  • Doctor auto-start on cold standalone — doctor checks now use CLI auto-start policy instead of suppressing on explicit ports
  • Doctor server-backed runtime detection — layered detection for server-backed runtime on older metadata.json without dolt_mode
  • Credential key location — moved from .beads/dolt/ to .beads/ to avoid creating ghost directories in shared-server mode
  • Colons in query values — unquoted identifiers like label=gt:merge-request now work without quoting
  • Init localhost warning — warns when server host defaults to localhost for remote Dolt setups
  • Repo-local Dolt server preservation — stale cleanup preserves servers from other repos
  • Backup restore UUID IDs — accepts UUID-format comment and event IDs during restore

[0.60.0] - 2026-03-12

Added

  • bd bootstrap with recovery actions — bd bootstrap now executes recovery actions directly instead of just printing advice, making first-run and repair workflows hands-free (GH#2438, GH#2372)
  • bd context command — new command for safe-first error guidance, surfacing relevant context when things go wrong (fix-merge PR #2496)
  • bd help --list and bd help --doc — CLI doc generation flags for producing machine-readable command listings and full documentation (GH#2527)
  • bd done alias — bd done is now an alias for bd close, aligning with Gas Town's session lifecycle vocabulary
  • bd done <id> <message> — last argument is treated as the close reason, removing the need for --comment
  • --design-file flag — read design documents from files instead of stdin for bd create workflows (PR #2524)
  • --destroy-token for safe re-init — enables non-interactive bd init re-initialization with explicit confirmation token (GH#2497)
  • GitHub Issues integration — tracker plugin for syncing GitHub Issues with beads (GH#2373)
  • Global PRIME.md fallback — ~/.config/beads/PRIME.md is used when no project-level PRIME.md exists (GH#2330)
  • Shared Dolt server mode — multi-repo and multi-agent setups can share a single Dolt server instance (GH#2416)
  • Ephemeral port allocation for Dolt — replaces hash-derived ports with OS-assigned ephemeral ports, eliminating port collisions entirely (GH#2415)
  • Workspace identity preflight — bd bootstrap validates workspace identity before any operations (GH#2438)
  • Epic close guards — prevents accidental closure of epics with open children; includes merge re-parenting and progress display
  • bd search searches external_ref — external references (GitHub issue URLs, Linear IDs, etc.) are now searchable (GH#2494)
  • JSON-aware error output — errors include structured JSON when --json is active, with JSONL schema validation and contract tests (GH#2499)
  • ai.api_key config — fallback for ANTHROPIC_API_KEY environment variable, stored in config.yaml
  • Embedded Dolt storage interfaces — new storage abstraction layer contributed by DoltHub (coffeegoddd) for future embedded Dolt work (PR #2427, PR #2428)
  • Community tools — added beads-web and claude-protocol to COMMUNITY_TOOLS.md; added community forks listing (PR #2457)

Changed

  • Dolt port allocation — switched from deterministic hash-derived ports to OS-assigned ephemeral ports; stored in repo-local state file rather than derived at runtime (GH#2415)
  • Auto-push state moved to local file — prevents metadata merge conflicts during bd dolt pull by keeping auto-push state out of the shared metadata (GH#2466)
  • YAML config prefix precedence — YAML config prefix now takes precedence over DB-stored prefix in shared-server mode (GH#2469)
  • Charm library upgrades — migrated glamour to v2 (charm.land/glamour/v2) and huh to v2 (charm.land/huh/v2) (GH#2471)
  • Daemon infrastructure fully removed — final cleanup of all remaining daemon references and idle monitor infrastructure (GH#2431, GH#2452)
  • Dead sync mode scaffolding removed — removed remaining sync mode code that was left behind after the Dolt migration (GH#2485)
  • 3-way merge engine remnants removed — cleaned up leftover code from the removed 3-way merge system (GH#2391)
  • Nix flake — uses go mod edit instead of sed for version patching (more reliable)

Fixed

  • Dolt journal corruption — KillStaleServers now runs inside flock to prevent concurrent server kills from corrupting the Dolt journal (GH#2430)
  • Dolt config corruption — uses explicit DOLT_ADD to prevent config corruption from stale working set state (GH#2455)
  • Dolt pull merge errors — auto-commits pending changes before pull to prevent merge errors (GH#2474)
  • Dolt pull transaction safety — wraps DOLT_PULL in explicit transaction for autocommit compatibility (GH#2501)
  • Dolt remote directory — bd dolt remote add/list/remove now operate on the correct directory (GH#2306, GH#2311)
  • Dolt endpoint drift — warns when auto-started server endpoint doesn't match config (GH#2399)
  • CLI remotes synced into SQL server — CLI-managed remotes are now synced into the SQL server on store open (GH#2315)
  • Server mode .beads/ creation — correctly creates .beads/ directory when using external BEADS_DOLT_* env vars (GH#2519)
  • Metadata merge conflicts — auto-resolves metadata merge conflicts during bd dolt pull (GH#2466)
  • Metadata replacement bug — --metadata now merges with existing metadata instead of replacing it (GH#2423)
  • Backup file detection — detects backup files on database-not-found and after bd init (GH#2327)
  • bd export crash — fixed crash on zero-time timestamps and stale bd sync refs (GH#2488, GH#2493)
  • bd children — now includes closed children by default (GH#2481)
  • bd doctor --fix — regenerates missing metadata.json (GH#2482); backfills project_id for pre-GH#2372 Dolt stores (GH#2490); blocked at Gas Town town root to prevent accidental damage (GH#2450)
  • bd doctor — detects stale .legacy hook sidecars (GH#2398); --clean --json now works correctly (GH#2438)
  • bd lint — accepts non-empty acceptance_criteria field without heading (GH#2468)
  • bd list --json — no longer ignored when --tree defaults to true (three separate fixes consolidated)
  • bd mol wisp — reports correct count when RootOnly=true
  • bd init — prevented from creating DB on another project's Dolt server (GH#2336); init guard allows fresh clones with committed metadata.json (GH#2433); data-dir blocked in server mode (GH#2438)
  • Init guard error message — removed dangerous --force suggestion
  • JSONL backups — preserve dependency metadata (GH#2487); restore denormalized JSONL without data loss (GH#2479)
  • Reparented children — excluded from molecule auto-close (GH#2480)
  • Worktree hooks — use absolute path for core.hooksPath so hooks work in worktrees (GH#2414)
  • Worktree .gitignore — skip append when parent directory pattern already covers the entry
  • Git hook shims — added timeout and graceful DB-missing handling
  • Stale lock files — follow redirect target when cleaning stale lock files
  • BEADS_DIR — stabilized paths from detached commit worktrees; prefer stable branch worktrees
  • BEADS_DOLT_PORT fallback — added fallback in GetDoltServerPort() (GH#2486)
  • PersistentPreRun flag checks — use Root().PersistentFlags() for all persistent flag checks (GH#2514)
  • Stale pinned flag — auto-cleared on status transition
  • Install script — detects WSL and MINGW/MSYS/CYGWIN environments (PR #2409)
  • Test stability — Dolt test suite eliminates container crashes and cascading failures; timezone-aware date assertions; same-type issues for blocks dependency tests

Performance

  • Close-time template subgraph loading — significantly faster epic close operations (GH#2458)

Documentation

  • Stale sync references purged — comprehensive cleanup of bd sync, bd import, --branch refs across CLI help, setup generators, and docs (GH#2522, GH#2435, GH#2442)
  • Stale migration ref fixed — bd migrate --to-dolt references replaced with bd init --from-jsonl (GH#2333)
  • QUICKSTART.md — removed stale --branch flag and SQLite references (GH#2522)
  • CI doc validation — added CI check to catch stale documentation references
  • Community tools — added beads-web and claude-protocol

Dependencies

  • github.com/dolthub/driver updated (multiple digests)
  • golang.org/x/sync v0.20.0
  • golang.org/x/sys v0.42.0
  • golang.org/x/term v0.41.0
  • OpenTelemetry Go monorepo v1.42.0
  • Testcontainers Go monorepo v0.41.0
  • Charm glamour v1.0.0 → v2.0.0
  • Charm huh v0.8 → v2.0.0

[0.59.0] - 2026-03-05

Added

  • bd list --tree default — tree view is now the default display mode for bd list (GH#2345)
  • OpenCode recipe — bd setup now includes an OpenCode integration recipe
  • Fresh clone detection — bd doctor detects fresh clones on Dolt server and guides through init (GH#2372)
  • Config-aware init guard — bd init distinguishes server-reachable from DB-exists, with better error messages (GH#2372)
  • Sync remote hint in errors — database-not-found errors now surface sync.git-remote config for easier troubleshooting
  • Nix flake modernization — updated flake for nixpkgs-25.11, fix-merge of PR #2314
  • Doctor warning suppression — suppress specific bd doctor warnings via config (GH#1095)
  • Community Tools — added fancypantalons/nvim-beads to COMMUNITY_TOOLS.md

Changed

  • 3-way merge config removed — ConflictStrategy, FieldStrategy, and all associated config/validation removed; Dolt handles merge natively (GH#2353)
  • Daemon infrastructure removed — the bd daemon has been fully removed; bd is now purely CLI-driven (w-bd-001)
  • Legacy daemon lock infrastructure removed — cleaned up stale lock files and related code
  • Backup git-push defaults to OFF — bd backup git push now requires explicit opt-in (GH#2363)
  • Test infrastructure modernized — migrated all tests from StartTestDoltServer to container-native API; readiness expanded to 5-state enum; crash detection via standalone functions (GH#2304)
  • ExtractPrefix consolidated — moved to types package for reuse across codebase

Fixed

  • Hook marker handling — bd doctor --fix repairs broken/orphaned hook markers; version added to END marker for future matching (GH#2344)
  • Legacy hook migration — warn on user-modified legacy hooks; handle backup sidecar files during migration
  • Dolt push/pull directory — use correct database subdirectory for CLI push/pull/fetch operations
  • Contributor auto-routing — add fallback routing for show/update/close commands (GH#2345)
  • Config-aware prefix detection — ResolvePartialID now respects config for prefix routing
  • Doctor command order — correct enrichFreshClone command sequencing
  • Batch IN-clause queries — prevent full table scans in all IN-clause queries (GH#2294)
  • Init port resolution — use DefaultConfig for port resolution during init (GH#2372)
  • Idle monitor — single-instance lock, port config isolation, project UUID verification (GH#2367, GH#2372)
  • Init data safety — prevent data destruction from misleading error messages (GH#2363, GH#2372)
  • Circuit breaker tuning — reduce cooldown to 5s and add active TCP health probe
  • Deterministic ordering — add ID tiebreaker to all ORDER BY clauses
  • Backup on feature branches — skip git commits during hook runs and on feature branches
  • Backup config routing — route backup.* keys to config.yaml so overrides are respected
  • Embedded Dolt — handle context canceled in cleanup, add SQL concurrency test
  • Tombstone entries — skip tombstone entries in bd init --from-jsonl
  • DerivePort — restore as standalone default in DefaultConfig; log fallback on collision
  • Cross-type blocking — align validation with cross-type blocking rules
  • Cross-prefix dep routing — extend to dep commands (GH#2296)
  • Dolt serialization retry — retry on serialization error in concurrent tests
  • CLI routing — gate on local remote availability; extend SQL push timeout (GH#2295)
  • Docker Hub network calls — skip when Dolt image not cached (GH#2277)
  • Wisp batch deletes — commit per batch in deleteWispBatch to avoid write timeout
  • Idle monitor watchdog — kills itself via Stop() to prevent zombie processes

Documentation

  • Audit and clean up stale sync mode documentation (w-bd-004)
  • Network/privacy section added to SECURITY.md
  • DerivePort birthday-problem collision recovery documented
  • Misleading examples corrected in QUICKSTART.md
  • GIT_INTEGRATION.md updated re deprecated beads-merge and Jujutsu
  • Circuit breaker and upgrade recovery troubleshooting docs
  • Duplicate nvim-beads entry fixed in COMMUNITY_TOOLS.md

[0.58.0] - 2026-03-02

Added

  • Persistent agent memory — bd remember, bd memories, bd recall, bd forget for knowledge that survives sessions and account rotations. Backed by the k/v store, auto-injected at bd prime time.
  • bd purge — delete closed ephemeral beads (wisps) to reclaim storage (GH#1692)
  • bd mol last-activity — show the most recent activity timestamp for any molecule (GH#1456)
  • bd show --current — show the active issue without specifying an ID (GH#2184)
  • bd doctor validate — Dolt-native conflict detection for data integrity checks (GH#2249)
  • bd init --backend — explicit backend selection with SQLite deprecation notice
  • --stdin flag — alias for --body-file - on bd create and bd update
  • bd preflight --check — aligned with CI checks for pre-push validation (GH#2230)
  • gofmt + golangci-lint pre-commit hook — automatic code quality enforcement (GH#2179)
  • JSONL-to-Dolt migration script — standalone bash script for pre-0.50 users (GH#2276)
  • bd doctor --fix — bridge pending hook migrations into automated repair
  • bd create-form --parent — create sub-issues with label inheritance

Fixed

  • Dolt CPU spikes — batch IN-clause queries in dependencies.go and computeBlockedIDs to prevent runaway CPU usage on large databases (GH#2294)
  • Dolt joinIter hangs — avoid hanging queries in GetReadyWork blocker computation
  • Stealth mode backup push — no-git-ops config now correctly prevents backup git push (GH#2290)
  • Stale DB connection crash — retry stale connection and preserve temp file in bd edit (GH#2267)
  • OSC escape leaks — prevent escape sequence pollution when third-party hook runners (lefthook, husky) call bd hooks run directly (GH#1303)
  • validateIssueIDPrefix — now checks allowed_prefixes config, unblocking convoy creation with multi-prefix databases (GH#1179)
  • Molecule steps in bd ready — remove stale SQLite-era filtering that hid mol steps (GH#1359)
  • bd blocked children — include children of blocked parents in output (GH#1495)
  • Same-type blocking validation — enforce that dependencies match allowed type pairs
  • Cross-prefix dependency routing — bd dep add/rm/blocks/tree now routes by prefix (GH#2285)
  • Cross-prefix dep existence check — skip target existence check for cross-prefix dependencies
  • Wisps table recreation — recreate dolt_ignore'd wisps tables on schema fast-path (GH#2271)
  • Missing wisps table — gracefully handle in epic/blocked queries (GH#2271)
  • Wisp dependency scanning — scan wisp_dependencies in getChildrenOfIssues and getChildrenWithParents
  • Cascade-delete wisp children — delete blocked step children during wisp GC
  • Batch wisp deletes — reduce Dolt commit pressure in DeleteIssues
  • Dolt push/pull hint — show helpful message when remote not found (GH#2118)
  • Git-protocol remote fallback — use CLI fallback for all git-protocol remotes, not just SSH (GH#2268)
  • dolt_data_dir persistence — strip absolute paths from metadata.json on save (GH#2251)
  • Dolt config port — consult config.yaml for dolt.port in DefaultConfig (GH#2073)
  • Missing dolt_database — detect and repair in metadata.json (GH#2160)
  • Dolt remote guards — guard remote subcommands against store init bypass (GH#2224)
  • bd search performance — avoid LIKE %% full-table scans
  • Worktree DB discovery — skip own .beads/ in separate-DB discovery (GH#2190)
  • Contributor routing — detect fork workflow, apply routing to list and ready
  • Pre-commit Go version mismatch — detect before golangci-lint panics
  • Viper defaults — skip in getRoutingConfigValue to unblock DB fallback
  • bd mol wisp create — default to root-only unless formula sets pour=true
  • Bare repo init — skip AGENTS.md generation in bare repositories
  • last_import_time refresh — prevent false staleness after write commands (GH#2255)
  • Hook marker detection — validate marker count and order; recognize marker-managed hooks (GH#2244, GH#2242)
  • Gitignore template — add missing dolt runtime files (GH#2256)
  • Pre-0.56 Dolt database upgrade — graceful recovery path (GH#2137)
  • Remotesapi port check — skip when no federation peers configured (GH#2273)
  • Doctor subsystem — fix panics, masked errors, missing rows.Err(), federation port, hardcoded paths; add 3 new tests
  • N+1 query patterns — replace with batch queries in deep.go and migration_validation.go
  • IsBlocked reporting — align blocker reporting with blocked-cache semantics (GH#1524)

Removed

  • Beads Classic SQLite backend — the SQLite storage layer and all migration infrastructure have been removed. Dolt is the only backend. (GH#87493ce)
  • go-sqlite3 dependency — no more CGO requirement; tests use Dolt testcontainers
  • Deprecated commands — removed dead SQLite-era code and legacy scripts

Documentation

  • Updated stale references to JSONL and SQLite architecture
  • Hook migration preview/apply workflow documentation

[0.57.0] - 2026-03-01

Added

  • bd doctor --agent mode — AI agent diagnostics for automated health checks (bd-6ud)
  • SSH push/pull fallback — dual-surface remote management (SQL + CLI) with automatic SSH detection (#2178)
  • Hook migration system — census and migration planning for git hooks with doctor signal (#2217)
  • Section markers for git hooks — hooks now use section markers for safer updates (GH#1380)
  • Auto-close molecule root — molecule root automatically closes when all steps complete (#2200)
  • CLI aliases — --comment alias for bd close, --yes/-y alias for bd mol burn
  • bd backup commands — init, sync, and restore for Dolt-native backups
  • JSONL backup & auto-backup — bd export command for JSONL backup, auto-enable when git remote exists
  • bd gc, bd compact, bd flatten — standalone beads lifecycle management
  • Circuit breaker — Dolt server connection circuit breaker to prevent agent hangs
  • Config-driven metadata schema — enforcement via metadata_schema config (GH#1416)
  • Metadata flags — --metadata for bd create, --set-metadata/--unset-metadata for bd update, metadata filtering in bd ready (GH#1406)
  • PreToolUse hook — blocks interactive cp/mv/rm prompts in agent workflows
  • Auto-update stale hooks — bd init detects and updates outdated hooks (GH#1466)
  • --root-only flag — for bd mol wisp create
  • hk support — hk.jdx.dev as supported git hook manager (#2182)
  • bd dolt remote — add, list, remove subcommands for Dolt remote management
  • Auto-push to Dolt remote — automatic push with 5-minute debounce after Dolt commits
  • dolt-data-dir config — custom dolt data directory (#2143)
  • --database flag for bd init — configure existing server database (#2102)
  • Per-worktree redirect — .beads/redirect overrides main repo .beads (#1995)
  • has_metadata_key — new predicate in bd query DSL (#1996)
  • MCP claim tool — atomic start-work for MCP integrations (#2071)
  • Auto-clean noms LOCK — stale Dolt noms LOCK file cleanup on startup (#2059)
  • Stale database detection — in bd doctor and bd dolt clean-databases
  • Self-managing Dolt server — port collision fallback, idle monitor, crash watchdog for standalone users (GH#2049, GH#2050)
  • bd dolt start/stop — server lifecycle management commands
  • Counter mode — issue_id_mode=counter for sequential issue IDs (#2013)
  • Auto-migration shim — SQLite to Dolt migration via sqlite3 CLI
  • Auto-migrate on first command — SQLite to Dolt auto-migration on first bd command
  • Linear Project sync — sync Linear projects to beads (#2022)
  • Label inheritance — child issues inherit parent labels on creation
  • Infra type routing — route agent/rig/role/message beads to wisps table
  • IsInfraType config-driven — via types.infra config key
  • bd init --stealth — sets no-git-ops: true in config (GH#2159)
  • Staleness check — read-only commands validate against JSONL freshness
  • CLI feedback titles — command feedback now shows issue titles (GH#1384)
  • OTel design docs — OpenTelemetry architecture and data model documentation (#2195)
  • Jira V2 API — support for Jira V2 API (#2088)

Fixed

  • Shadow database prevention — bd no longer runs CREATE DATABASE IF NOT EXISTS unconditionally; errors with clear diagnostic instead of creating shadow databases
  • Auto-start suppressed with explicit server port — prevents shadow database creation when configured server is temporarily down
  • Phantom catalog entries — respect existing dolt_database config in init and migrate paths (#2051)
  • information_schema resilience — replace with SHOW COLUMNS/SHOW TABLES to avoid stale catalog crashes
  • bd doctor --server — fix crash on phantom entries
  • Reparented child display — reparented child no longer appears under old parent (#2001)
  • Homebrew publishing — use brews instead of homebrew_casks for formula publishing; auto-publish tap on release
  • NULL created_by crash — handle NULL in issue scan to prevent bd duplicates crash (#2198)
  • Dolt UNION crash — avoid SELECT * UNION that crashes Dolt on column count mismatch
  • SHOW DATABASES LIKE — replace with exact-match iteration for Dolt compatibility
  • Dolt port resolution — doctor checks and all commands use hash-derived port instead of hardcoded 3307 (#2148)
  • dolt_data_dir respect — doctor checks use configured data directory (#2181)
  • Wisps table defaults — add default values to NOT NULL TEXT columns (bd-5nn)
  • AUTO_INCREMENT reset — reset counters after DOLT_PULL (#2135)
  • bd init --stealth — correctly sets no-git-ops: true in config (GH#2159)
  • Skip autopush on read-only — prevent writes on read-only commands (GH#2191)
  • Sandbox guard — restore sandbox guard in maybeAutoPush (#2192)
  • Mol notes field — propagate notes field from formula steps to issues (#2155)
  • Graph children layer — lift children to parent's layer in graph layout (GH#1748)
  • DefaultInfraTypes mutation — unexport to prevent inconsistency (#2193)
  • Conditional-blocks readiness — evaluate conditional-blocks deps in computeBlockedIDs() (#2128)
  • Dolt connection timeouts — prevent agent hangs with connection timeouts
  • Migration safety — verify DB target, deduplicate, spot-check data during migration (#2156)
  • Ephemeral read-path fallback — fall through to issues table when wisps table missing (#2161)
  • Ephemeral explicit-ID lookup — complete routing across all storage layers (GH#2053)
  • waits-for deps in bd blocked — include in output (GH#2043)
  • Reopen/undefer no-op status — report correct status (GH#2037)
  • Post-create metadata — commit deps and labels to Dolt after create (GH#2009)
  • Status validation — validate against built-in + custom statuses on update (#2021)
  • Empty title/label rejection — reject on update/label add (GH#1994)
  • Non-zero exit on failure — exit non-zero when all close/update attempts fail (GH#2014)
  • JSON metadata validation — validate on creation path (#2032)
  • Wisp field validation — parity with issues in createWisp (#2032)
  • Empty comment rejection — reject whitespace-only comment text
  • bd stale validation — reject non-positive --days flag
  • bd defer past date warning — warn when --until date is in the past
  • Doctor non-git-repo — gracefully handle in fingerprint and role checks
  • Doctor cruft cleaning — clean cruft in redirect-expected dirs without redirect file (#2194)
  • Doctor gitignore — follow redirect when fixing outdated .beads/.gitignore (#2078)
  • Doctor maintenance — migrate maintenance checks to Dolt (#2146)
  • Dolt server management — Gas Town guardrails, one server per town, kill-before-start (#2089)
  • Migration 007 resilience — handle schema evolution differences (#2168)
  • Windows compatibility — Makefile ldflags fix, connectex syscall handling, doltserver filesystem heuristic (#2165, #2171, #2092)
  • Batch SQL queries — prevent query explosion with IN-clause batching (#2108)
  • Routing write divergence — check local store before prefix routing (#2106)
  • Hook outdated detection — unify between CheckGitHooks and hooksNeedUpdate (#2105)
  • Linear status sync — include stateId in push updates (#2144)
  • Dep type overwrites — prevent silent dependency type overwrites on dep add
  • Prefix sanitization — sanitize hyphens to underscores for SQL database names (GH#2142)
  • Issue prefix normalization — prevent double-hyphen bead IDs
  • SQLite # in URI paths — escape to prevent truncation (#2150)
  • bd dep traversal — traverse blocks-type edges in dep tree and graph commands (#2151)
  • bd dolt push/pull/commit — allow to initialize store (GH#2042)
  • bd init --from-jsonl — restore for server mode
  • Doctor stale verification — improved verification after --fix (#2077)
  • Upsert on insert — prevent duplicate primary key errors (GH#2061)
  • Database version auto-migrate — auto-migrate on CLI upgrade for all commands
  • Wisp double-prefixing — stop wisp ID double-prefixing and add wisp fallback to ResolvePartialID
  • Dolt DOLT_COMMIT transaction — commit SQL tx before DOLT_COMMIT to persist wisp data
  • OSC escape leaks — prevent in git hooks (GH#1303)
  • bd ready --json — include labels, deps, and parent in output
  • Git upstream warning — suppress for repos with no remotes (#2116)
  • bd init prefix sanitization — sanitize auto-detected prefix for MySQL database names
  • Rejected flag-like args — reject flag-like positional args in bd create
  • Backup streaming — stream backup exports to disk, reduce allocations
  • Version scripts — update for section markers, suppress unused param lint

Performance

  • Test parallelization — Dolt storage tests 3.5x faster, protocol tests 3x faster
  • Branch-per-test isolation — shared DB with branch-per-test for doctor (44s → ~12s), cmd/bd, and protocol tests
  • Testcontainers migration — test Dolt server uses testcontainers instead of binary spawning
  • Consolidated test suites — messaging and dep test suites merged for efficiency
  • Add phantom catalog detection to bd doctor
  • Fix bd doctor --server crash on phantom entries (replace INFORMATION_SCHEMA.SCHEMATA with SHOW DATABASES)

[0.56.1] - 2026-02-23

Fixed

  • Release CI — remove verify-cgo.sh post-hook from darwin and freebsd builds which intentionally use CGO_ENABLED=0 (cross-compilation without CGO)

[0.56.0] - 2026-02-23

Removed

  • Embedded Dolt mode — beads now requires a running Dolt SQL server. The embedded Dolt driver (dolthub/driver) and all CGO build-tag bifurcation have been removed. Binary size drops from 168MB to ~41MB. Use bd dolt start or dolt sql-server to run the server.
  • SQLite ephemeral store — ephemeral issues (wisps) now live in Dolt-backed wisps table with dolt_ignore, replacing the separate SQLite database. Run bd migrate wisps to migrate existing data.
  • JSONL sync pipeline — the entire JSONL-based sync system (bd sync, git-portable mode, belt-and-suspenders mode) has been removed. Dolt-native push/pull via git remotes is the only sync mechanism. bd sync is now a deprecated no-op.
  • JSONL bootstrap — clone initialization uses Dolt clone only; JSONL bootstrap path removed.
  • JSONL plumbing — cross-rig JSONL flush, JSONL recovery in doctor, and JSONL-based restore have been removed.

Added

  • Metadata query support — bd list, bd search, and bd query now support --metadata-field key=value and has_metadata_key filters (#1908)
  • Metadata visibility — bd show and bd list --long display metadata in human-readable format (#1905)
  • Wisps table — ephemeral issues stored in dedicated Dolt-backed table with dolt_ignore for compaction-friendly lifecycle
  • bd migrate wisps — migration command for SQLite-to-Dolt ephemeral data
  • Batch auto-commit mode — reduces Dolt commit bloat by batching writes with SIGTERM/SIGHUP flush
  • --agents-template flag — bd init now supports named AGENTS.md templates via embed.FS
  • Mux setup recipe — bd setup mux with layered AGENTS.md and managed hooks
  • Standalone formula execution — bd mol wisp supports expansion formulas (#1903)
  • Sentinel errors — ErrNotFound, ErrNotInitialized, ErrPrefixMismatch for cleaner error handling
  • --skip-prefix-validation — bd import flag for legacy data migration
  • Protocol invariant test suite — data integrity and blocking semantics regression tests (#1910)
  • OpenTelemetry instrumentation — opt-in OTLP tracing for hooks and storage operations (#1940)
  • Transaction infrastructure — RunInTransaction with commit messages, isolation, retry, and batch wrapping for Dolt concurrency

Fixed

  • bd ready ordering — respect SortPolicy and preserve result ordering (#1883)
  • waits-for readiness — bd ready and molecule analysis now correctly handle waits-for dependencies (#1900)
  • Dependency tree parent ID — populate ParentID in tree output and show [BLOCKED] for root (#1992)
  • Parent-child display — bd list now separates parent-child deps from blocking deps (#1948)
  • Hook shim templates — use bd hooks run instead of nonexistent bd hook command
  • Cross-expansion dependencies — bd mol cook propagates deps across formula expansions (#1901)
  • Wisp auto-close — wisp root automatically closes after squash (#1898)
  • Dolt server writes — commit via execContext when server runs with --no-auto-commit
  • Metadata normalization — normalize metadata and waiters in UpdateIssue
  • Batch import — persist labels, comments, and deps during import
  • Noms LOCK detection — use flock probe instead of file existence (#1960)
  • Doctor backend awareness — deep validation checks configured backend, not directory presence
  • Doctor federation — use configured database name in federation and health checks (#1904, #1924, #1925)
  • Doctor orphan deps — exclude external: deps from orphan check (#1593)
  • Jira sync — use correct API v3 /search/jql endpoint (#1953)
  • Plugin dep order — correct inverted bd dep add argument order in plugin docs (#1928)
  • Wisp routing — fix multiple ephemeral store routing gaps for create, read, promote, and gc
  • Prime output — remove stale --from-main flag reference
  • bd list resolved blockers — treat missing/unreachable blockers as resolved (#1884)
  • Wisp search parity — add ~15 missing filter fields to searchWisps
  • Wisp label hydration — hydrate labels in getWispsByIDs for search results
  • Query nil guard — prevent panic in GetBlockedIssues and GetEpicsEligibleForClosure
  • Issue prefix clobber — guard SetConfig to prevent overwrite when rigs share a Dolt database
  • Atomic bond/squash — bd mol bond, bd mol squash, and bd mol cook now run in single transactions (bd-wvplu, bd-4kgbq)
  • bd ready parent filter — pass --parent filter to GetReadyWork/GetBlockedIssues and propagate blocked status to children (#2009, #1495)
  • bd list sort/limit — --limit now applies after --sort; trim whitespace in bd edit (#1237, #1234)
  • Doctor lock false positive — use flock probe for noms LOCK, remove stale bd sync references (#1981, #2007)
  • Repo sync consistency — cross-prefix hydration and close guard consistency (#1945, #1524)
  • DOLT_COMMIT in CRUD — all write operations now produce Dolt commits for history tracking
  • Double JSON encoding — fix daemon-mode RPC response unmarshaling across show, dep, label, reopen (bd-4ec8)
  • G304 gosec finding — annotate os.ReadFile in tips.go with #nosec for constructed paths (bd-8g8)
  • Stale daemon references — remove obsolete bd daemon references from all documentation (#1982)

Performance

  • Binary size — 168MB → ~41MB (dropped dolthub/driver and wazero WASM runtime)
  • Linux/Windows startup — eliminated 2-second wazero JIT compilation penalty on every invocation
  • Test suite — doctor tests 89s → 28s; shared DB pattern across cmd/bd suites
  • Test isolation — dolt package and cmd/bd tests now isolated from production Dolt server (bd-2lf6)
  • N+1 queries — batch dependency/label/comment queries with per-invocation caching (#1874)

[0.55.4] - 2026-02-20

Fixed

  • Release CI: FreeBSD build — disable CGO for FreeBSD target; zig sysroot lacks stdlib.h

[0.55.3] - 2026-02-20

Fixed

  • Release CI: macOS builds — disable CGO for darwin targets; zig sysroot lacks macOS frameworks (CoreFoundation, Security) and libresolv; pure-Go build via gms_pure_go + netgo tags

[0.55.2] - 2026-02-20

Fixed

  • Release CI: libresolv — strip -lresolv entirely from zig wrappers instead of replacing with -lresolv.9 (zig sysroot has neither); macOS builds use netgo tag so libresolv is not needed
  • Release formula — added rate limit warnings to prevent burning GitHub API quota during CI wait

[0.55.1] - 2026-02-20

Fixed

  • Release workflow YAML — heredoc in zig wrapper step broke YAML literal block parsing; replaced with echo commands
  • Version consistency — marketplace.json was not bumped in v0.55.0
  • Go formatting — ran gofmt on 8 files
  • Lint — fixed unused parameter and dead code in config.go

[0.55.0] - 2026-02-20

Fixed

  • Release CI: zig upgrade — upgrade zig 0.13.0 → 0.14.0 to fix AccessDenied bug in cross-compilation (ziglang/zig#20689)
  • Release CI: macOS libresolv — work around zig 0.14.0 libresolv resolution by using versioned -lresolv.9 (ziglang/zig#16674)
  • Release CI: Android ARM64 — disable CGO for Android target (server mode only)
  • 5 pre-existing test failures — resolve test failures and Dolt panic (bd-iqsw6v)
  • Doctor daemon references — remove daemon references for post-daemon architecture

Removed

  • ~5K lines dead code — classic cleanup of obsolete sync pipeline remnants

[0.54.0] - 2026-02-18

Fixed

  • Release CI — fix zig cross-compilation cache race by serializing goreleaser builds (--parallelism 1)
  • Android ARM64 build — disable CGO for Android target (server mode only)
  • .zig-cache/ in .gitignore — prevent goreleaser dirty-git false positive

[0.53.0] - 2026-02-18

Added

  • Dolt-in-Git sync — native Dolt push/pull via git remotes replaces the entire JSONL sync pipeline; BootstrapFromGitRemote(), HasRemote(), sync.git-remote config key
  • bd dolt start / bd dolt stop — explicit Dolt server management commands (#1813)
  • bd dolt commit — desire-path ergonomics for committing Dolt data (#1812)
  • Server mode without CGO — OpenFromConfig() exported for server-mode aware store opening (#1805)
  • Hosted Dolt support — TLS, authentication, and explicit branch configuration
  • bd mol wisp gc --closed — bulk purge of closed wisps
  • --no-parent flag for bd list — filter out child issues
  • bd ready pretty format — improved default output with priority sort, truncation footer, and parent epic context
  • Dolt compaction methods — bd compact support for Dolt databases
  • Storage interface — Storage interface decouples from concrete DoltStore (bd-l3o)
  • Lock health diagnostics — bd doctor detects and reports dolt-access.lock and noms LOCK issues
  • Codecov integration — component-based coverage tracking in CI

Fixed

  • Pre-commit deadlock on embedded Dolt — resolve hook deadlock (#1841, #1843)
  • bd doctor --fix hang — run fixes in-process instead of spawning subprocess (#1850)
  • Dolt lock errors — surface lock errors with actionable guidance instead of silent empty results (#1816)
  • BEADS_DIR config loading — respect BEADS_DIR when loading config.yaml (#1854)
  • Dolt Unknown database retry — retry on Unknown database after CREATE DATABASE (#1851, #1852)
  • Windows Expand-Archive — qualify to avoid Pscx module conflict (#1826)
  • bd init git repo check — ensures git repo exists and runs clean diagnostics
  • SaveConfigValue — preserves existing file contents (#1816)
  • molecule core type — treat molecule as core issue type (#1866)
  • Formula VarDef — distinguish "no default" from default="" (#1837)
  • Worktree test isolation — harden metadata.json isolation (bd-la2cl)
  • Goroutine leak — replace time.Sleep with waitFor in tests (#1822)

Removed

  • JSONL sync-branch pipeline — deleted internal/syncbranch/ (5,720 lines), snapshot_manager, deletion_tracking, doctor sync-branch checks/fixes, and all associated tests (~11,000 lines total)
  • Daemon infrastructure — removed lockfile activity signal and orchestrator (bd-9u8zd)
  • 3-way merge engine remnants — removed remaining merge code (bd-aswyy)
  • Dead stubs — CheckBackendMigration, NeedsJSONL, and other no-ops (bd-23nlq)
  • Dead SQLite-only bd repair command

Documentation

  • BeadHub — added to COMMUNITY_TOOLS.md
  • beads-compound plugin — added to Claude Code Orchestration section (#1814)
  • Stale reference cleanup — removed daemon/SQLite/JSONL/tombstone references from docs and comments

[0.52.0] - 2026-02-16

Added

  • bd ready --include-ephemeral — new flag to include ephemeral issues in ready work results

Fixed

  • Doctor redirect target resolution — fix path resolution for redirect targets (#1803)
  • Dolt directory creation guard — skip dolt directory creation when not in server mode (#1800)
  • Tilde expansion in core.hooksPath on Windows — properly expand ~ in git hook paths (#1798)
  • Worktree redirect path resolution — resolve redirect paths from worktree root, not .beads/ dir (#1791)
  • Block rename-prefix in worktrees — prevent rename-prefix from running in git worktrees to avoid JSONL staleness (#1792)
  • CI lint warnings — resolve gosec G204 and unparam lint warnings (#1790)

Removed

  • Dead git-portable sync functions — remove unused gitCommitBeadsDir and gitPush (#1793)

[0.51.0] - 2026-02-16

Changed

  • Dolt-native cleanup (dolt-1s40) — massive 8-phase refactoring to remove all legacy SQLite/JSONL/daemon infrastructure:
    • Phase 2: Remove daemon compat stub and --no-daemon flag
    • Phase 3: Remove 3-way merge engine
    • Phase 4: Remove tombstone/soft-delete system
    • Phase 5: Remove JSONL sync layer
    • Phase 6: Remove SQLite backend entirely
    • Phase 7: Remove storage factory, memory backend, and provider abstraction; replace with direct Dolt calls
    • Phase 8: CLI & config cleanup — remove remaining SQLite/daemon remnants
  • Post-cleanup pruning — removed 12 dead/duplicate DoltStore methods (650 lines), 5 legacy doctor checks (1,457 lines), internal/beads type aliases, stale no-op stubs and dead functions
  • bd sync is now a no-op — sync layer removed as Dolt handles persistence directly
  • Dolt-native API renamed — legacy storage API consolidated to direct DoltStore calls
  • Server watchdog — added Dolt server health monitoring

Fixed

  • Dolt config test corruption — tests calling setDoltConfig() used FindBeadsDir() which in worktree environments returned the rig's .beads/ instead of the test's temp dir; fixed with t.Setenv("BEADS_DIR", beadsDir) (#1780)
  • Dolt embedded mode database name — read from metadata.json instead of hardcoding (#bd-pqn2e)
  • Batch DeleteIssues hang — queries batched to prevent hang on large ID sets, with correctness hardening and comprehensive tests (#1770)
  • bd mol current step readiness — use analyzeMoleculeParallel for accurate step readiness (#1786, #1276)
  • bd doctor AccessLock — migrate Dolt health checks to shared AccessLock connection with error-path tests and CloseWithTimeout documentation (#1780)
  • bd doctor --yes — honor --yes flag for repo fingerprint auto-fix (#1782)
  • bd doctor sqliteConnString pragmas — restore pragmas dropped by Phase 8 cleanup
  • GetReadyWork type filtering — exclude workflow/identity types from ready work results (#1774)
  • bd create dependency direction — swap dependency direction for explicit blocks: prefix (#1742)
  • bd dolt seed custom types — seed types.custom in default config (#1733)
  • message built-in type — add message as recognized built-in issue type (#1742)
  • Routing redirect resolution — lock relative redirect resolution for routed lookups (#1751)
  • Schema init skip — skip schema initialization when already at current version (#1765)
  • Test race conditions — serialize dolt.New() and stdio redirection to eliminate test races
  • deleteBatch signature mismatch — fix signature and unused types import
  • gofmt — formatting fixes across 24+ files
  • Bounded lock contention — enforce bounded lock retry behavior in syncbranch read/write tests (#1777)
  • Malformed test names — fix test names and remove obsolete SQLite-backend doctor tests

Performance

  • CASCADE deletes — leverage SQL CASCADE to cut deletion queries by 60%
  • Schema init — skip when already at current version, reducing startup overhead

Documentation

  • SQLite → Dolt migration — updated 10+ documentation files replacing stale SQLite references with Dolt (ARCHITECTURE.md, FAQ, TROUBLESHOOTING, WORKTREES, INTERNALS, and more)
  • Deleted deprecated docs — removed EXTENDING.md and MULTI_REPO_HYDRATION.md (SQLite-era, no longer applicable)
  • INSTALLING.md — revised for mise installation methods (#1756)
  • CLAUDE.md — fix stale file references and SQLite mentions
  • Contributor docs — updated stale SQLite references to Dolt
  • Dolt migrate command — fix typo (#1779)

[0.50.3] - 2026-02-15

Added

  • SyncEngine hook system — PullHooks (GenerateID, TransformIssue, ShouldImport) and PushHooks (FormatDescription, ContentEqual, ShouldPush, BuildStateCache/ResolveState) allow tracker-specific behaviors without modifying the engine
  • Jira native integration — extracted into internal/jira/ package with REST API v3 client, ADF document conversion, field mapping, and full test coverage
  • Tracker plugin registry — tracker.Register() + init() pattern for auto-discovery of tracker implementations (Linear, GitLab, Jira)

Changed

  • SyncEngine refactor — all three tracker CLIs (Linear, GitLab, Jira) now use the shared tracker.Engine for Pull→Detect→Resolve→Push orchestration, eliminating ~800 lines of duplicated sync code
  • Tracker adapters inlined — moved from internal/tracker/adapters/{linear,gitlab} into internal/{linear,gitlab} as self-contained packages with tracker.go and fieldmapper.go

Fixed

  • Jira State mapping bug — removed stale *StatusField pointer assignment in jiraToTrackerIssue that could cause incorrect status mapping when Priority was set but Status was nil
  • CI: Windows build — renamed test_wait_helper.go to _test.go suffix so non-test builds don't try to resolve test-only symbols
  • CI: gofmt — fixed formatting across 14 files
  • Formula variable validation — use Required field for formula variable validation
  • bd slot routing — added routing and label-based agent check to slot commands

Performance

  • Test suite — replaced ~60 git init subprocess calls with cached template copy, reducing test setup overhead

[0.50.2] - 2026-02-14

Added

  • SQLite-to-Dolt migration nudge — rate-limited hint (once per 24h) on any bd command for SQLite users, plus bd doctor backend migration check (#1706)
  • Dolt audit log for bd dolt config set changes

Fixed

  • bd doctor performance — replaced O(n) full-table scans with SQL aggregation queries, reducing doctor runtime from 130s to 6s on large databases (#1706)
  • Dolt metadata writes — reliable bd_version, repo_id, clone_id persistence with init-time verification, doctor fix, and migrate extension (#1741)
  • FindBeadsDir worktree boundary — stopped directory walk from escaping sibling worktree boundaries (#1731)
  • bd doctor plugin install command — corrected command shown in doctor output (#1732)
  • Backend descriptions — updated bd backend list and help text to show Dolt as the default (not SQLite) since v0.50

[0.50.1] - 2026-02-14

Fixed

  • CI: goreleaser failure — removed dead internal/rpc import from test stubs that broke go mod tidy
  • CI: Windows smoke test — use --backend sqlite and temp dir to avoid Dolt dependency and repo .beads/ conflicts
  • CI: lint errors — updated golangci-lint v2 exclusions for fmt.Fprintf errcheck, gosec G104/G304
  • CI: gofmt — fixed formatting in init_test.go, main.go, sync_git_remote_test.go

Added

  • Plugin-based issue tracker framework with Linear and GitLab adapters (#1150)

[0.50.0] - 2026-02-14

Added

  • Dolt is now the default backend for new bd init projects. Existing SQLite projects are unaffected. BEADS_DB env var auto-detects SQLite.
  • bd graph visualization overhaul — terminal-native horizontal DAG as default view, plus DOT and interactive HTML export formats. Old box view available via --box (bd-9de)
  • bd sql command for raw SQL access to the underlying database. Supports table, JSON, and CSV output
  • bd help --all to dump complete command reference (#1699)
  • decision built-in issue type with aliases and help strings
  • Cross-database dependency resolution via prefix routes in bd show, bd graph, and bd blocked — external deps display actual title, status, and priority (bd-k0pfm)
  • bd show --watch flag for auto-refreshing display on file changes with fsnotify debounce
  • bd vc commit --stdin for reading multi-line commit messages from stdin
  • BD_BRANCH env var for branch-per-polecat write isolation in Dolt
  • BD_NAME env var for multi-instance help text identity
  • bd doctor artifact cleanup — --check=artifacts detects stale JSONL, WAL/SHM, backup files. Works with --clean and --fix
  • bd doctor Claude Code integration checks — detect malformed settings.json, verify hook completeness, detect legacy MCP tool references
  • bd doctor output grouped by category with per-category pass counts
  • Dolt corruption recovery via bd doctor --fix — backs up corrupted .dolt-data/, reinitializes fresh Dolt database, auto-imports from JSONL backup
  • Frictionless bd init — better test-issue detection, non-interactive stdin detection, version tracking at init time, cancellable prompts, actionable error hints via FatalErrorWithHint
  • Mise installation documented in INSTALLING.md

Changed

  • Default backend is Dolt — bd init without --backend now creates Dolt databases. Existing projects with no explicit backend in metadata.json continue to use SQLite (backward compatible)
  • Removed daemon/RPC subsystem — internal daemon, RPC layer, and internal/rpc/ package deleted (~19,663 lines). All commands use direct embedded database access
  • Removed JSONL sync layer — internal/importer/, markDirtyAndScheduleFlush(), and daemonClient != nil branches eliminated (~7,634 lines)
  • Stripped dirty tracking from Storage interface — MarkIssueDirty subsystem no longer part of the storage contract
  • bd graph default view changed from vertical box layout to horizontal DAG
  • Schema version downgrade protection — CLI refuses to run against a database written by a newer version
  • Upgrade notification only triggers for actual version upgrades (no longer fires on downgrades/branch switches)
  • resolve-conflicts defaults to canonical issues.jsonl
  • Batch GetLabelsForIssues replaces N+1 GetLabels calls in list operations
  • Consolidated redirect resolution into single FollowRedirect function
  • Backend-agnostic refactoring — compact, migrate, dep, sync, findReplies, and wisp/mol commands use Storage interface instead of SQLite type assertions

Fixed

  • bd close enforces gate satisfaction before closing machine-checkable gates (gh:pr, gh:run, timer, bead). --force bypasses
  • bd show exits non-zero when issue not found (previously exited 0 with empty JSON output)
  • bd list closed-blocker filtering — closed issues no longer shown as blockers in annotations
  • bd list --parent includes dotted-ID children — matches both explicit parent links and X.* naming convention
  • bd ready deferred-parent propagation — excludes children of deferred parents from ready work
  • Dolt joinIter panic prevented by replacing IN/EXISTS subqueries with Go-level filtering
  • Embedded Dolt self-deadlock in git hooks and bd migrate — added to noDbCommands
  • Dolt rename FK violation — disable FK checks during rename operations
  • Dolt GetIssuesByLabel deadlock — fixed connection pool exhaustion with MaxOpenConns=1
  • XSS vulnerabilities in graph HTML export (bd-67uzz)
  • Redirect path resolution — resolve from parent of .beads dir, not .beads itself
  • bd migrate --to-dolt sets sync.mode=dolt-native and uses config-based backend
  • bd doctor backend-agnostic checks — uses storage factory instead of raw SQLite; routes.jsonl false positive fix; dolt_mode server respected
  • bd show extra newline before first issue header removed
  • --type flag help text shows only base types
  • bd prime prompt optimized to avoid spurious "creating issue without description" warning
  • Linuxbrew ICU paths supported in Makefile
  • Windows build recipe fixed (BINARY conditional moved inside targets)
  • go install fixed by removing local replace directive
  • Nix flake build and gosec lint errors resolved
  • Duplicate Cobra command registration replaced with Aliases
  • Homebrew upgrade command corrected in upgrading docs

Documentation

  • Added Dolt GitHub link to README
  • Added beadsmap and Beadbox to community tools
  • Corrected Homebrew upgrade command
  • Added Mise installation instructions to INSTALLING.md

[0.49.6] - 2026-02-08

Reverted

  • Embedded Dolt mode restored - The v0.49.5 removal was premature; embedded mode is still needed in Beads (only Gas Town should be server-only). Restores dolthub/driver, advisory flock, embedded connector lifecycle, CGO build tags, and vendored go-icu-regex

[0.49.5] - 2026-02-08

Added

  • bd search content and null-check filters - Filter issues by description content and metadata presence (--has, --no) (bd-au0.5)
  • bd promote command - Promote wisps to persistent beads (gt-7mqd.9)
  • bd todo command - Lightweight task management for quick personal tracking
  • bd find-duplicates command - AI-powered duplicate issue detection (bd-581b80b3)
  • bd validate command - Data-integrity health checks, integrated into bd doctor --check=validate (bd-e108)
  • Dolt fail-fast TCP check - Quick connectivity check before MySQL protocol initialization
  • Windows PE version info - Embedded version metadata reduces antivirus false positives (bd-t4u1)
  • Centralized AI model config - ai.model config key with DefaultAIModel() helper
  • Newsletter generator - Automated narrative release summaries (#1197)

Changed

  • Embedded Dolt mode removed - Server-only Dolt connections now; embedded driver fully removed (bd-esqfe). Windows Dolt backend connects via MySQL protocol. Note: reverted post-release — see [Unreleased]
  • bd init defaults to chaining hooks - No longer prompts; chains by default (bd-bxha)
  • Homebrew formula name - bd doctor now correctly suggests brew upgrade beads instead of brew upgrade bd
  • Doctor output - Summary-first layout with improved formatting (bd-4qfb)
  • JSON output standardized - Consistent JSON format across all commands (bd-au0.7)
  • Go toolchain - Downgraded from 1.25.7 to 1.25.5 for Nix compatibility. Note: back to 1.25.7 after embedded Dolt restore — see [Unreleased]

Fixed

  • Security: SQL injection prevention - SQL identifier validation for dynamic table and database names (dolt-test-46a)
  • Security: path traversal - Fixed path traversal in export handler and command injection in import
  • Security: CVE-2025-68121 - Go toolchain upgrade for TLS runtime fix
  • RPC: empty mutation events - Pass issueID via closure to prevent zero-value IDs in label/dep operations (dolt-test-rgh)
  • RPC: server shutdown - Drain in-flight requests before closing storage; proper response-write-then-shutdown pattern
  • RPC: nil storage guards - Added to handleCommentList, handleCommentAdd, and health handler
  • RPC: daemon socket path - Use ShortSocketPath for status/health/metrics commands
  • Daemon: YAML config parsing - Recognize both hyphenated and underscored daemon config keys (e.g., auto-sync and auto_sync)
  • Daemon: stderr redirect - Removed unsafe global os.Stderr replacement during import
  • Daemon: daemon stop suggestion - bd doctor fix suggestions now include workspace path argument
  • Doctor: role check - Falls back to database config for users who set role via bd config set
  • Doctor: gastown detection - Auto-detect gastown mode when routes.jsonl exists
  • Doctor: routing mode - Accept explicit as valid routing mode alongside auto/maintainer/contributor
  • Doctor: Dolt conflicts query - Use correct table column name (not table_name)
  • Sync: conflict resolution - Apply --ours/--theirs strategy correctly
  • Sync: tombstone export - Include tombstones in auto-flush full export
  • Merge: panic prevention - Use strings.HasPrefix to prevent panic on short error messages
  • Import: label sync - Remove DB labels absent from JSONL during import
  • Import: issueEqual comparison - Include dependencies and comments in comparison
  • Export: dirty flag preservation - Prevent stdout export from clearing dirty flags and auto-flush state
  • Export: temp file collisions - Use unique temp filenames in writeMergedStateToJSONL
  • Migration: bounds checks - Add bounds checks for slice/string access in migrate.go
  • Migration: daemon check - Check for running daemon before dolt migration
  • Migration: error logging - Log warnings for label and dependency import errors
  • Migration: legacy spec_id - Fix doctor --fix migration for legacy spec_id
  • SQLite: WAL retry deadlock - Make Close() idempotent to prevent deadlock (bd-4ri)
  • SQLite: BUSY retry - Use SQLITE_BUSY retry logic for all BEGIN IMMEDIATE calls (bd-ola6)
  • SQLite: JSONL locking - Add exclusive lock to flushToJSONLWithState to prevent race conditions
  • Dolt: connection retry - Make "connection refused" retryable in retry logic
  • Dolt: cross-rig contamination - Use prefix-based database names; detect contamination in post-migration check
  • Dolt: port consistency - Use port 3307 consistently for server connections
  • Config: beads.role - Write to git config instead of SQLite
  • Config: viper guard - Log when viper uninitialized in getConfigList
  • Config: BD_BACKEND blocked - Block env var to prevent data fragmentation (bd-hevyw)
  • bd list output - Separate parent-child from blocks display; --all disables default limit
  • bd dep list routing - Fix cross-rig routing from town root (bd-ciouf)
  • Cross-prefix ID resolution - Support multi-repo scenarios (GH#1513)
  • Staleness checks - Added to export, graph, history, gate list commands
  • Gitignore template - Add .jsonl.lock to template and requiredPatterns
  • MCP plugin - Remove output_schema workaround for FastMCP 2.14.4
  • Nix/Windows CI - Unbroken: removed orphaned pure_go_windows.go, updated flake.lock, fixed vendorHash

Documentation

  • Messaging system documentation (messaging.md)
  • Issue metadata field and reserved key prefixes
  • Docker trick for updating flake.lock without nix
  • Community Tools: JetBrains beads-manager plugin, Beadspace desktop app
  • Fix ready docs and external-ref documentation (#1523, #1339)

[0.49.4] - 2026-02-05

Added

  • Label pattern and regex filtering - New --label-pattern (glob) and --label-regex flags for bd list and bd ready. Filter issues by label patterns like tech-* or regex like tech-(debt|legacy) (#1491)
  • Simple query language - Complex filtering via a simple query syntax for bd list
  • beads command alias - Installer now creates a beads alias alongside bd
  • spec_id field - New field for linking issues to specification documents (#1372)
  • Wisp type field - wisp_type column for TTL-based compaction of ephemeral molecules
  • Dolt schema migration runner - Automated schema migrations for Dolt backend
  • Dolt migration validation in bd doctor - Checks for Dolt migration health
  • Agent-managed Dolt upgrade path - bd migrate support for Dolt upgrades
  • --metadata flag for bd update - Update issue metadata from CLI via JSON (#1417, bd-0vud2)
  • UpdateIssue metadata via json.RawMessage - Programmatic metadata updates (#1417)
  • BD_SKIP_ACCESS_LOCK env var - For Dolt lock testing scenarios
  • Orphan branch support for sync-branch migration - Cleaner sync branch setup
  • config.local.yaml support - Local configuration overrides via .beads/config.local.yaml
  • Makefile up-to-date check - make install now verifies repo is current before building

Changed

  • bd ready now shows only open issues - Excludes in_progress, matching bd list --ready behavior. Shows work that is truly available to claim
  • release.sh is now a molecule gateway - Creates a release wisp instead of running a batch script
  • Merge driver embeds types.Issue - Prevents field drift between merge driver and core types (#1481)

Fixed

  • JSONL file locking - Prevents race conditions in concurrent writes and incremental exports
  • WAL checkpoint robustness - Retry logic for SQLite WAL checkpointing
  • Merge driver field drift - Driver now preserves all issue fields including spec_id, metadata, and dependencies (#1480, #1481, #1482)
  • Atomic bd claim - Compare-and-swap semantics prevent race conditions (#1474)
  • BEADS_DIR routing - Respects BEADS_DIR over prefix routing in show/close/update commands
  • Dolt-native sync fixes - Disabled JSONL sync exports, auto-flush, and imports in dolt-native mode to prevent conflicts
  • Daemon auto-import - Proper JSONL import on startup with hash update; check changes AFTER pulling
  • Windows build - Dolt now builds on Windows via pure-Go regex; prebuilt releases in installer
  • Cross-rig dependency resolution - Uses factory.NewFromConfig for proper backend detection
  • Dolt lock contention - Advisory flock prevents zombie processes; scoped migration helpers
  • Dolt connection stability - Retry logic for transient errors, timeout on embedded close, fix for nested query bad connections
  • Dolt FK constraint - Removed FK constraint on depends_on_id for external references
  • bd list --tree deduplication - Children no longer appear multiple times
  • bd show relates-to display - Correctly displays and deduplicates RELATED section
  • bd cleanup --hard - Always prunes tombstones immediately
  • Formula TOML vars - Accept simple string vars in [vars] section
  • config.yaml fallback - Custom types work in daemon mode
  • Content-based merge for daemon pulls - Prevents spurious conflicts (GH#1358)
  • TOML snake_case tags - Proper serialization of formula struct fields
  • Worktree bare repo support - Use GetGitCommonDir for worktree creation
  • routes.jsonl corruption - Excluded from FindJSONLInDir to prevent import errors
  • Compaction safety logic - Restored accidentally removed safety checks
  • Structured logging - Replaced custom daemonLogger with *slog.Logger

Documentation

  • Non-interactive shell command guidance for agents
  • Makefile CGO settings and DOLT.md env vars
  • Atomic claim feature in quick reference
  • ICU build deps and installer hints
  • Community Tools: beads-sdk and Beads Task-Issue Tracker

[0.49.3] - 2026-01-31

Changed

  • Embedded Dolt is now the default - Server mode is opt-in via dolt_mode: "server" in metadata.json or BEADS_DOLT_SERVER_MODE=1 env var

Fixed

  • Dolt split-brain root cause eliminated (B1+B2) - DatabasePath() now always resolves to .beads/dolt/ for dolt backend regardless of stale database field values; bootstrapEmbeddedDolt() blocks JSONL auto-import in dolt-native sync mode to prevent silent rogue database creation
  • CGO/ICU build fix - Makefile and test.sh detect Homebrew's keg-only icu4c and export CGO flags so dolt's go-icu-regex dependency links correctly on macOS
  • Dolt mergeJoinIter panic - Eliminated three-table joins that triggered panics on type-filtered queries; added guard against nil pointer panic during auto-import in dolt-native mode
  • Template variable extraction - Filter Handlebars keywords (if, each, unless, with) from extractVariables (#1411)

[0.49.2] - 2026-01-31

Added

  • GitLab backend for bidirectional issue sync - Full GitLab integration for syncing beads issues with GitLab

    • REST API client wrapper with pagination, rate limiting, and context cancellation
    • Bidirectional sync: pull from GitLab, push to GitLab with conflict detection
    • bd gitlab sync, bd gitlab status, bd gitlab projects subcommands
    • GitLab-to-beads mapping with path-based IDs and collision handling
    • Conflict resolution strategies for concurrent edits
    • Integration and unit test suites
  • Key-value store - New bd kv subcommand for persistent key-value storage

    • bd kv get, bd kv set, bd kv delete, bd kv list commands
    • Key validation and proper exit codes on missing keys
  • Per-issue JSON metadata field - Optional metadata field on issues (SQLite + Dolt) (#1407)

  • Events JSONL export - Opt-in event audit trail export via events-export config

  • Role configuration - Explicit role configuration via git for agent and user roles

    • bd init interactive contributor role prompt
    • UserRole detection in RepoContext
    • Agent roles configurable via config (bd-hx8w)
  • Dolt improvements

    • Dolt-specific diagnostics and performance profiling in bd doctor
    • Auto-detect Dolt server and enable server mode during bd init
    • Env var overrides for Dolt server connection settings
    • bd backend and bd sync mode subcommands for inspecting storage config
  • CLI improvements

    • comment_count in issue JSON views (list, ready, search)
    • Comment timestamps with time display and --local-time flag
    • Hint to view step instructions in bd mol current output (#1403)
    • Resolve external refs in bd dep list for cross-rig dependencies
    • Inline blocking dependency display in bd list output
  • Jira sync - pull_prefix and push_prefix config options for flexible prefix mapping

  • Sync - Push permission error detection during sync

Changed

  • Separation of concerns - Removed Gas Town-specific code from beads core

    • Removed Gas Town role detection from hooks
    • Removed --role-type flag from bd create
    • Removed Gas Town actor validation from validation layer
    • Generalized RoleType to be project-agnostic
  • Refactoring

    • Storage factory for backend-aware database access
    • Compact accepts interface instead of concrete SQLiteStorage
    • RPC layer no longer imports sqlite directly
    • JSONL deletion markers now processed during import

Fixed

  • GitLab fixes - Multiple rounds of code review improvements for merge readiness

    • Conflict detection before push with SyncContext
    • Pagination limits and context cancellation
    • Error handling and type safety improvements
  • Dolt backend fixes

    • Graceful server-to-embedded fallback when Dolt server unreachable
    • Detect and clean stale Dolt lock files
    • Skip JSONL staleness check in dolt-native mode
    • Backend detection in all bd doctor fix functions
    • Use GetDoltDatabase() instead of raw config for server mode
    • Factory-based storage creation for backend awareness
    • Fix for stale JSONL in dolt-native mode
    • SQLite not created when Dolt fails in dolt-native mode
    • Revert Dolt lock cleanup workaround; fix embedded Dolt open via driver retries (#1389)
  • Sync fixes

    • Set GIT_DIR and GIT_WORK_TREE in GitCmd for worktree support
    • Normalize JSONL path in CommitToSyncBranch for sync-branch mode
    • Use store.Path() for database mtime update
    • Ensure sync-branch worktree exists on fresh clone (#1349)
    • Skip sync when source and destination JSONL paths are identical (#1367)
    • Respect dolt-native mode in daemon_sync_branch.go
    • Respect dolt-native mode in JSONL export paths
    • Auto-stage JSONL files after flush in pre-push hook
  • Other fixes

    • Ignore undeclared handlebars in formula description text (#1394)
    • Exclude ephemeral issues from Linear sync push (#1397)
    • Allow event type in CreateIssue validation (#1398)
    • Update MCP Stats model to match bd stats --json output (#1392)
    • Handle EINVAL from chmod on Unix sockets in containers (#1399)
    • Correct error message for --gated flag (#1391)
    • Check rows.Err() in SQLite GetAllEventsSince
    • Exclude tombstones from external_ref uniqueness validation
    • Handle nil pointer dereference in bd restore with invalid issue ID
    • Make daemon start idempotent (#1334)
    • YAML config support for daemon auto-sync settings (#1294)
    • Use IsValidWithCustom for --type validation (#1356)
    • Skip SQLite-specific checks in doctor when using Dolt backend
    • Tombstone wins over closed in merge conflict resolution

Documentation

  • GitHub Copilot integration guide with cross-platform MCP paths (#1348)
  • KV store CLI documentation
  • Updated COMMUNITY_TOOLS.md
  • Added export-state/ to gitignore template

[0.49.1] - 2026-01-25

Added

  • Dolt backend now fully supported - The Dolt storage backend has been extensively tested and is ready for community evaluation

    • Note: Dolt is not enabled by default. We encourage users to try it out and report feedback!
    • Auto-commit on write commands with explicit commit authors (#1267)
    • Server mode for multi-client access - enables multiple processes to share a Dolt database
    • bd doctor --server flag for Dolt server mode health checks
    • Server mode configuration in metadata.json schema
    • Comprehensive test suite for Dolt storage backend (#1299)
    • Lock retry and stale lock cleanup for operational reliability (#1260)
    • Adaptive ID length instead of hardcoded 6 chars
    • See docs/DOLT.md for setup guide
  • New command flags

    • bd activity --details/-d for full issue information in activity feed (#1317)
    • bd export --id and --parent filters for targeted exports (#1292)
    • bd update --append-notes flag to append to existing notes (#1304)
    • bd update --ephemeral and --persistent flags (#1263)
    • bd show --id flag for IDs that look like flags
  • Doctor improvements

    • --server flag for Dolt server mode health checks
    • Stale closed issues check is now configurable (#1291)
  • Community tools - Added beads-kanban-ui, beads-orchestration (#1255), and abacus (#1262)

  • Newsletter generator - Automated narrative summaries of releases

Changed

  • Homebrew installation - Now uses core tap for beads installation (#1261)
  • GitHub Actions - Upgraded for Node 24 compatibility (#1307, #1308)

Fixed

  • Dolt backend fixes

    • Skip JSONL checks in pre-push hook for dolt-native mode
    • Add ID generation to transaction CreateIssue
    • Use capabilities check instead of blanket Dolt block in daemon
    • Recognize shim hooks as Dolt-compatible
    • Add hook compatibility check and migration warning
    • Skip JSONL sync for Dolt backend
    • Refuse daemon startup with deprecated --start flag for Dolt backend
    • Add YAML fallback for custom types/statuses
    • Prevent daemon startup and fix routing same-dir check
    • Proper server mode support for routing and storage
    • Skip daemon auto-start for all Dolt backends
  • Daemon fixes

    • Prevent stack overflow on empty database path (#1288, #1313)
    • Add sync-branch guard to daemon code paths (#1271)
  • List command fixes

    • Optimize bd list --json to fetch only needed dependencies (#1316)
    • Prevent nil pointer panic in watch mode with daemon (#1324)
    • Populate dependencies field in JSON output (#1296, #1300)
  • Import/export fixes

    • Support custom issue types during import (#1322)
    • Populate export_hashes after successful export (#1278, #1286)
    • Use transaction for handleRename in upsertIssuesTx (#1287)
  • SQLite fixes

    • Use BEGIN IMMEDIATE without retry loop (#1272)
    • Add retry logic to transaction entry points (#1272)
    • Use withTx helper for remaining transaction entry points (#1276)
  • Other fixes

    • bd gate add-waiter now functions correctly (#1265)
    • Respect BEADS_DIR environment variable in init (#1273)
    • Find molecules attached to hooked issues (#1302)
    • Separate parent-child deps from blocking deps in bd show (#1293)
    • PrefixOverride now respected in transaction storage layer (#1257)
    • Prime command defaults to beads, avoids TodoWrite/TaskCreate (#1289)
    • Build: make install now properly builds, signs, and installs to ~/.local/bin

Documentation

  • Dolt backend guide - Comprehensive docs/DOLT.md guide (#1310)
    • Federation section with quick start and topologies
  • Articles collection - Created ARTICLES.md for articles and tutorials about Beads (#1306)

[0.49.0] - 2026-01-21

Added

  • Dolt federation for multi-repo sync - Peer-to-peer issue synchronization across repositories

    • bd federation sync command for syncing with configured peers
    • bd federation status shows connection and sync state
    • SQL-server mode for daemon (--federation flag) enables peer connections
    • SQL user authentication for secure peer-to-peer sync
    • Doctor federation health checks validate connectivity and sync state
    • See docs/FEDERATION.md for setup guide
  • SQLite to Dolt migration - Migrate existing repos to version-controlled storage

    • bd migrate dolt converts SQLite database to Dolt backend
    • Preserves full issue history during migration
    • Automatic JSONL bootstrap for routes and interactions
  • New commands

    • bd children <id> - Display child issues for a parent
    • bd rename <old-id> <new-id> - Rename issue IDs
    • bd view - Alias for bd show command (#1249)
    • bd config validate - Validate sync configuration
    • bd mol seed --patrol - Seed patrol molecules (#1149)
  • Sync improvements

    • Per-field merge strategies for fine-grained conflict resolution
    • Interactive conflict resolution with --manual strategy
    • Incremental export for large repositories (performance improvement)
    • sync.mode configuration to control sync behavior
  • CLI conveniences

    • -m flag as alias for --description in bd create
    • --type and --exclude-type flags for Linear sync filtering (#1205)
    • --tree --parent combination for hierarchical display (#1211)
  • Jujutsu (jj) version control support - Beads now works with jj repositories

    • Hook support for jj operations
    • Proper detection of jj-managed repositories
  • Codex CLI setup recipe - Automated setup for OpenAI Codex CLI integration (#1243)

  • Real-time activity feed - Uses fsnotify for instant updates

  • Per-worktree export state tracking - Each worktree maintains independent sync state

  • Automatic multi-repo hydration in bd init --contributor - Routing and hydration now configured together

    • bd init --contributor automatically adds planning repo to repos.additional
    • Routed issues appear in bd list immediately after setup
    • No manual bd repo add required for contributor workflow
  • Doctor improvements

    • Routing+hydration mismatch detection - warns when routing configured without hydration
    • Hydrated repo daemon check - ensures JSONL stays fresh in additional repos
    • Patrol pollution detection and fix
    • --gastown flag for Gas Town-specific checks (#1162)
    • Federation health checks
  • Nix shell completions - Baked into default package (#1229)

Changed

  • Auto-routing disabled by default - Explicit opt-in required (#1177)

    • Prevents unexpected cross-repo routing
    • Enable with routing.mode: auto in config.yaml
  • Gas Town types removed from core - Beads core no longer includes Gas Town-specific types

    • Types like patrol, convoy moved to Gas Town configuration
    • Cleaner separation between beads and Gas Town
    • Ongoing cleanup tracked in bd-741si, bd-7nd6t, bd-31ajf

Fixed

  • Daemon zombie state after database file replacement - Improved reconnection resilience (#1213)

    • Health checks now properly detect and handle database file replacements
    • Prevents "sql: database is closed" errors
  • Routed issues invisible in bd list (split-brain bug) - Auto-flush JSONL after routing (#1251)

    • bd create now flushes JSONL immediately in target repo
    • Hydration now sees new issues immediately
  • WSL2 Docker Desktop compatibility - Detect bind mounts and disable WAL mode (#1224)

    • Prevents database corruption on Docker Desktop file systems
  • Daemon stack overflow - Prevent recursion in handleStaleLock (#1238)

  • Molecule steps excluded from bd ready - Filter out non-actionable items (#1246)

  • Tree ordering stabilization - Consistent --tree output (#1228)

  • Cross-repo orphan detection - Honor --db flag (#1200)

  • Custom types with daemon - Show custom types when daemon is running (#1216)

  • Dolt backend fixes

    • Parse timestamps from TEXT columns correctly
    • Single-process mode enforcement (daemon/autostart disabled) (#1221)
    • Read-only daemon commands now work
    • Init/daemon/doctor integration fixes (#1218)
  • Redirect handling - Follow redirect when creating database

  • Auto-import tombstone handling - Auto-correct deleted status to tombstone (#1231)

  • Routing store cleanup - Close original store before replacing (#1215)

  • Worktree redirect paths - Correct path computation in bd worktree create (#1217)

  • Sync-branch worktree - Use worktree for --full --no-pull (#1183)

  • Custom types from config - Load types.custom during init auto-import (#1226)

Documentation

  • Federation setup guide - Added docs/FEDERATION.md

    • Peer configuration and authentication
    • Sync workflow documentation
  • Daemon troubleshooting guide - Added docs/TROUBLESHOOTING.md and docs/DAEMON.md

    • Common daemon issues and solutions
    • Database replacement handling
  • Multi-repo hydration guide - Added comprehensive section to docs/ROUTING.md

    • Explains hydration requirement when using routing
    • Troubleshooting guide for common issues
  • Contributor vs maintainer setup - Added to README.md

    • Clarifies when to use bd init --contributor
    • Documents role configuration options

[0.48.0] - 2026-01-17

Added

  • VersionedStorage interface - Abstract storage layer with history/diff/branch operations

    • Enables pluggable backends (SQLite, Dolt) with unified API
    • Supports time-travel queries and branching semantics
  • bd sync command specification - Formalized sync workflow implementation

    • Clearer separation between export, import, and merge phases
  • bd types command - List valid issue types (#1102)

    • Shows all available types with descriptions
  • "enhancement" type alias - Alternative name for "feature" type

    • Matches GitHub issue label conventions
  • bd close -m flag - Alias for --reason (git commit convention)

    • More intuitive for git users: bd close <id> -m "reason"
  • RepoContext API - Centralized git operations context (#1102)

    • Consistent git directory handling across codebase
  • Dolt backend improvements (WIP)

    • Automatic bootstrap from JSONL on first access
    • Git hook infrastructure for Dolt operations
    • bd compact --dolt flag for Dolt garbage collection

Fixed

  • Doctor sync branch health check - Removed destructive --fix behavior (GH#1062)

    • No longer warns about expected source file differences
    • Prevents accidental sync branch history destruction
  • Duplicate merge target selection - Use combined weight (GH#1022)

    • Considers both dependents and dependencies when choosing merge target
    • Prevents closing issues with children/dependencies
  • Worktree exclude paths - Use --git-common-dir for correct paths (GH#1053)

    • Fixes bd init --stealth in git worktrees
  • Daemon git.author config - Apply configured author to sync commits

    • Respects git.author config and BD_GIT_AUTHOR env var
  • Hook chaining preservation - Prevent --chain from destroying original hooks (#1120)

    • Backs up existing hooks before chaining
  • Sync routed prefixes - Allow routed prefixes in import validation

    • Fixes multi-prefix workflow issues
  • Windows CGO-free builds - Enable building without CGO (#1117)

    • Improved Windows compatibility
  • Shell completions without database - Work without .beads/ (#1118)

    • Completions function in non-beads directories
  • Timestamp normalization - Normalize to UTC for validation (#1123)

    • Prevents timezone-related validation failures
  • Symlinked .beads directories - Correct routing for symlinks (#1112)

    • Proper resolution of symlinked beads directories
  • Nil pointer in ResolvePartialID - Prevent panic (#1132)

    • Guards against nil pointer dereference
  • Orphaned dependencies on delete - Mark dependents dirty (#1130)

    • Prevents orphan deps in JSONL after issue deletion
  • Git hooks in worktrees - Fix hook execution (#1126)

    • Hooks now work correctly in linked worktrees

Documentation

  • CLI skill reference - Synced with v0.47.1 commands

    • Updated command documentation in claude-plugin
  • AGENTS.md fixes - Removed references to nonexistent CLAUDE.md

    • Corrected cross-references in documentation

[0.47.2] - 2026-01-14

Added

  • Dolt backend (experimental) - Version-controlled issue storage with time-travel

    • bd init --backend=dolt enables Dolt-based storage
    • Full version history for issues with branch/merge semantics
    • See docs/DOLT.md for migration guide
  • bd show --children flag - Display child issues inline with parent

    • Shows hierarchical structure in issue details
  • Comprehensive NixOS support - Improved Nix flake and home-manager integration

    • Better daemon handling in Nix environments
    • Updated flake.nix with proper dependencies

Changed

  • Release workflow modernized - bump-version.sh replaced with molecule pointer
    • Use bd mol wisp beads-release --var version=X.Y.Z for releases
    • New scripts/update-versions.sh for quick local version bumps

Fixed

  • Redirect + sync-branch incompatibility - bd sync works correctly in redirected repos (bd-wayc3)

    • Worktree git status failures resolved
    • Proper handling of .beads/redirect during sync operations
  • Doctor project-level settings - Detects plugins/hooks/MCP in .claude/settings.json

    • No longer misses project-scoped configurations
  • Contributor routing - bd init --contributor correctly sets up routing (#1088)

    • Fork workflows now properly configure sync.remote=upstream

Documentation

  • EXTENDING.md deprecated - Custom SQLite tables approach deprecated for Dolt migration
    • External tools pattern recommended for integrations

[0.47.1] - 2026-01-12

Added

  • bd list --ready flag - Show only issues with no blockers (bd-ihu31)

    • Filters to issues that are immediately actionable
    • Equivalent to bd ready but integrated into list command
  • Markdown rendering in comments - Comment text now renders Markdown (#1019)

    • Enhanced readability for formatted notes and descriptions

Changed

  • Release formula improvements - Updated beads-release formula with v0.47.0 learnings
    • Better gate handling and step organization

Fixed

  • Nil pointer in wisp create - Prevent panic when creating wisps (mol)

    • Fixed nil pointer dereference in molecule creation
  • Route prefix for rig issues - Use correct prefix when creating issues in rigs (#1028)

    • Issues created in rigs now use the proper routing prefix
  • Duplicate merge target selection - Prefer issues with children/deps (GH#1022)

    • Better heuristics for choosing merge target in duplicate detection
  • SQLite cache rebuild - Rebuild blocked_issues_cache after rename-prefix (GH#1016)

    • Ensures cache consistency after prefix changes
  • Doctor JSONL check - Exclude sync_base.jsonl from multiple files check (#1021)

    • Reduces false positives in doctor diagnostics
  • Merge struct completeness - Add QualityScore field to merge Issue struct

    • Ensures all fields preserved during merge operations
  • MCP custom types - Support custom issue types and statuses in MCP (#1023)

    • MCP server now handles non-built-in types correctly
  • Hyphenated prefix validation - Support hyphens in ValidateIDFormat (#1013)

    • Prefixes like my-project- now validate correctly
  • Git worktree initialization - Prevent bd init inside git worktrees (#1026)

    • Avoids configuration issues when initializing in worktree directories

Documentation

  • bd reset clarification - Document command behavior and workarounds (GH#922)
    • Clearer guidance on reset command usage

[0.47.0] - 2026-01-11

Added

  • Pull-first sync with 3-way merge - Major sync improvement (#918)

    • Reconciles local changes with remote updates before pushing
    • Field-level conflict merging reduces manual intervention
    • Base state tracking for better change detection
  • bd resolve-conflicts command - Resolve JSONL merge conflict markers (bd-7e7ddffa)

    • Mechanical mode uses updated_at timestamps for deterministic resolution
    • Closed status wins over open, higher priority wins
    • Notes concatenated, dependencies unioned
    • Dry-run mode and JSON output for agent integration
  • bd create --dry-run - Preview issue creation without side effects (bd-0hi7)

    • Shows what would be created in human-readable or JSON format
    • Works with --rig/--prefix flags
  • Gate auto-discovery - Auto-discover workflow run ID in bd gate check (bd-fbkd)

    • Queries GitHub directly when await_id is a workflow name hint
    • ZFC-compliant: takes most recent run deterministically
  • Linear project filter - linear.project_id config for sync (#938)

    • Fetch only issues from a specific project instead of all team issues
  • bd ready --gated - Gate-resume discovery for molecules (bd-lhalq)

    • Find molecules waiting on gates for automatic resumption
  • Multi-repo custom types - Trust and discover types across repositories (bd-62g22, bd-9ji4z)

    • bd doctor discovers custom types from multiple repos
    • Non-built-in types trusted during hydration
  • Visual UX improvements - Enhanced display for list tree, graph, and show commands

    • Better formatting and readability
  • Stale database handling - AllowStale option in List API (bd-dpkdm)

    • Read-only commands auto-import on stale DB (#977, #982)
    • Cold-start bootstrap for read commands
  • Batch molecule operations - bd mol burn supports multiple molecules (feat(mol))

  • Redirect health checks - bd doctor validates redirect configurations

  • Schema extensions - New fields for HOP integration

    • crystallizes column in sqlite storage
    • attests edge type for skill attestations
    • owner field for human attribution
    • actor fallback includes git user.name (#994)

Fixed

  • Daemon mode completeness - Several daemon mode gaps closed (GH#952)

    • --due and --defer flags now work in daemon mode (#953)
    • bd dep add/remove --json returns proper JSON output (#961)
    • DeferUntil field parsed correctly in daemon handleCreate (#950)
    • Silence deprecation warnings in --json mode (#1039a691)
  • Sync robustness

    • Canonicalize dbPath to fix filepath.Rel errors (GH#959, #960)
    • Validate custom types in batch issue creation (#943)
    • Force-add .beads in worktree for contributor mode (#947)
    • Initialize store after daemon disconnect (GH#984)
    • sync --import-only works when daemon was connected
  • Windows fixes

    • Infinite loop in findLocalBeadsDir/findOriginalBeadsDir (GH#996)
    • bd init no longer hangs when not in a git repo (#991)
    • Daemon stop/kill uses proper Windows API (GH#992)
    • SQLite uses DELETE mode on WSL2 Windows filesystem (GH#920)
  • Daemon socket handling - Long workspace paths now work (GH#1001, #1008)

    • Socket path shortening for deep directory structures
    • Relocate daemon socket for deep paths
  • Prevent data corruption

    • FK constraint failures on batch/concurrent issue creation (GH#956)
    • Prevent closing issues with open blockers (GH#962)
    • Nil pointer panic in dep --json mode (GH#998)
  • Doctor improvements

    • Recognize bd shims when external manager config exists (GH#946)
    • Detect lefthook jobs syntax (GH#981)
    • Add .sync.lock and sync_base.jsonl to gitignore (#980)
  • Prime command - Use flush-only workflow when no git remote configured (#940)

  • Install safety - Stop existing daemons before binary replacement (#945)

  • Git hooks - Add --no-daemon to sync commands to prevent inline import failures (#948)

  • Linear sync - Use project_id when creating issues via sync --push (GH#973, #1012)

  • Team wizard - Validate sync.branch in wizard and migrate commands (GH#923)

  • Worktree fixes

    • Skip beads restore when directory is redirected (bd-lmqhe)
    • Migrate sync works in git worktree environments (#970)
  • Misc fixes

    • bd edit parses EDITOR with args (GH#987)
    • Use SearchIssues for ID resolution (GH#942)
    • Respect hierarchy.max-depth config setting (GH#995, #997)
    • Add timeout to daemon request context to prevent hangs
    • Avoid null values in Claude settings hooks (GH#955)
    • Restore Gas Town types (agent, role, rig, convoy, slot) (GH#941)
    • Add TypeRig constant and IsBuiltIn method (GH#1002)

Changed

  • Daemon CLI refactor - Consolidated to subcommands with semantic styling (#1006)

  • Release formula refactor - Bump script broken into individual version-update steps (bd-a854)

    • More durable: can resume from specific step if interrupted
    • Better visibility in activity feed

Documentation

  • Add lazybeads (Bubble Tea TUI by @codegangsta) to community tools (#951)
  • Fix bd quickstart link to database extension documentation (#939)

[0.46.0] - 2026-01-06

Added

  • Custom type support - Configure custom issue types in beads config.yaml (bd-649s)

    • Define project-specific types beyond the built-in set
    • Types persist across sync and export
  • Gas Town types extraction - Core Gas Town types moved into beads package (bd-i54l)

    • Enables beads to understand rig identities and agent workflows
    • Foundation for deeper Gas Town integration

Fixed

  • Gate workflow discovery - Handle workflow name hints in gh:run gate discovery (bd-m8ew)
    • Better matching of GitHub Actions workflow runs
    • Consolidated numeric ID handling

[0.45.0] - 2026-01-06

Added

  • Dynamic shell completions - Tab completion for issue IDs in bash/zsh/fish (#935)

    • Optimized prefix filtering for faster completion
    • Added completions to more commands
  • Android/Termux support - Native ARM64 binaries for Android (#887)

  • Deep pre-commit framework integration - bd doctor checks pre-commit hook configs (bd-28r5)

  • Rig identity bead type - New rig bead type for Gas Town rig tracking (gt-zmznh)

  • --filter-parent alias - Alternative to --parent in bd list (bd-3p4u)

  • Unified auto-sync config - Simpler daemon configuration for agent workflows (#904)

  • BD_SOCKET env var - Test isolation for daemon socket paths (#914)

Fixed

  • External hook manager detection - bd doctor now detects lefthook, husky, pre-commit, and other hook managers

    • Checks if external managers have bd hooks run integration configured
    • Reports which hooks have bd integration vs which are missing
    • bd doctor --fix uses --chain flag when external managers detected to preserve existing hooks
    • Supports YAML, TOML, and JSON config formats for lefthook
    • Detects active manager from git hooks when multiple managers present
  • Init branch persistence - --branch flag now correctly persists to config.yaml (#934)

  • Worktree resolution - Resolve worktrees by name from git registry (#921)

  • Sync with redirect - Use inline import for --import-only with .beads/redirect (bd-ysal)

    • Handle .beads/redirect in git status checks (bd-arjb)
    • Persist sync branch to yaml and database (GH#909)
    • Atomic export and force-push detection (bd-3bhl, bd-4hh5)
  • Doctor improvements

    • Recognize lowercase 's' skip-worktree flag (#931)
    • Improve messaging for detection-only hook managers (bd-par1)
    • Align duplicate detection with bd duplicates (bd-sali)
    • Query metadata table instead of config for last_import_time (#916)
  • Update prefix routing - bd update now routes like bd show (bd-618f)

  • Formula phase - beads-release formula marked as vapor phase (gt-mjsjv)

  • Lint fixes - Address gosec, misspell, and unparam warnings

  • CI improvements - Windows smoke tests, timeouts, golangci-lint exclusions

Changed

  • Test refactoring - Replace testify with stdlib in daemon tests (#936)

Documentation

  • Add bun installation method to INSTALLING.md (#930)
  • Add Parade app to community tools list

[0.44.0] - 2026-01-04

Added

  • Recipe-based setup architecture - Major refactor of bd init (bd-i3ed)

    • Modular, composable setup via recipes
    • Cleaner separation of init concerns
  • Gate evaluation system - Phases 2-4

    • bd gate check for timer and GitHub gate evaluation (GH#884)
    • bd gate discover for auto-discovery of gh:run await_id (bd-z6kw)
    • bd gate add-waiter and bd gate show for phase handoff
    • Cross-rig bead gate support
    • Gate-aware beads-release.formula v2 (bd-r24e)
    • Merge-slot gate for serialized conflict resolution
  • Dependency command improvements

    • --blocks shorthand flag for natural dependency syntax (GH#884)
    • --blocked-by and --depends-on flag aliases (bd-09kt)
  • Multi-prefix support - allowed_prefixes config option (#881)

    • Configure multiple valid prefixes per rig
    • Enables flexible issue routing
  • Sync improvements

    • Detect uncommitted JSONL changes before sync (GH#885)
    • bd doctor sync divergence check for JSONL/SQLite/git
    • BD_DEBUG_SYNC env for protection debugging
  • PRIME.md override - Workflow customization (GH#876)

    • Custom prime output per project
  • Daemon config - daemon.auto_* settings in config.yaml (GH#871)

  • Compound visualization - bd mol show displays compound structure (bd-iw4z)

  • /handoff skill - Session cycling slash command (bd-xwvo)

Fixed

  • bd ready now shows in_progress issues (#894)

    • Previously filtered out your active work
  • bd show displays external_ref field in text output (#899)

  • macOS case-insensitive path handling (GH#880)

    • Worktree validation, daemon paths, git operations
    • Canonicalize path case for consistency
  • Sync metadata timing - Finalize after commit, not push (GH#885)

    • Defer SQLite metadata updates until after git commit
    • Prevents sync state corruption
  • Sparse checkout isolation - Prevent config leaking to main repo (GH#886)

    • Disable sparse checkout on main repo after worktree creation
  • close_reason preserved during merge/sync (GH#891)

  • Parent hub contamination during bd init (GH#896)

  • NoDb mode - Set cmdCtx.StoreActive correctly (GH#897)

  • Event storm prevention when gitRefsPath is empty (#883)

  • Doctor improvements

    • Detect status mismatches between DB and JSONL (GH#885)
    • Detect missing git repo, improve daemon startup message (#890)
    • Skip JSONL tracking checks in sync-branch mode (GH#858)
  • bd rename-prefix - Sync JSONL before and after (#893)

  • sync.remote config now respected in bd sync (GH#872)

  • Snapshot protection made timestamp-aware (GH#865)

  • Hyphenated rig names supported in agent IDs (GH#854, GH#868)

  • Cross-repo agent routing (#864)

  • Config key normalization - GetYamlConfig matches SetYamlConfig (#874)

  • Daemon startup failure - Propagate reason to user (GH#863)

  • Molecule variable substitution in root bead title/desc

  • Submodule detection - Correct main repo root detection (#849)

  • bd merge output sorted by issue id (#859)

  • Auto-create agent bead when bd agent state called on non-existent agent

  • Windows zip extraction - Add retry logic for npm install

  • Cycle detection runs in daemon mode for --blocks flag

  • bd slot set cross-beads prefix routing (bd-hmeb)

  • git status noise - Hide issues.jsonl when sync.branch configured (GH#870)

Changed

  • Skill names updated from /bd-* to /beads:* (#862)

Performance

  • Batch external dep checks by project (bd-687v)

Internal

  • Extract warnIfCyclesExist helper
  • Gate field parsing and creation tests
  • Sync unit tests for gitHasUncommittedBeadsChanges
  • SQLite cache rebuild benchmarks
  • Repository guards in deployment workflows

[0.43.0] - 2026-01-02

Added

  • Step.Gate evaluation - Phase 1: Human Gates

    • Gate steps that require human approval before proceeding
    • Foundation for workflow control points
  • bd lint command - Template validation

    • Validate issue templates against schema
    • --validate flag also added to bd create
  • bd ready --pretty - Formatted output

    • Human-friendly display of ready work

Fixed

  • Cross-rig routing - bd close and bd update now support cross-rig operations via prefix routing
  • Agent ID validation - Now accepts any rig prefix (GH#827)
  • bd sync in bare repo worktrees - Fixed exit 128 error (GH#827)
  • bd --no-db dep tree - Now shows complete tree (GH#836)
  • .beads/last-touched - Restored to gitignore template (GH#838)

[0.42.0] - 2025-12-30

Added

  • llms.txt standard support - AI agent discoverability (#784)

    • Standard endpoint for AI agents to discover project info
    • Helps LLM-powered tools understand your project
  • bd preflight command - PR readiness checks

    • Static checklist for pre-commit/pre-push verification
    • Phase 1 implementation with test, lint, and version checks planned
  • --claim flag for bd update - Work queue semantics

    • Atomic claim-and-assign for multi-agent coordination
    • Prevents race conditions in work assignment
  • bd state and bd set-state - Label-based state management

    • Helper commands for operational state patterns
    • Documents labels-as-state pattern
  • bd activity --town - Cross-rig activity feed

    • Aggregated view across all rigs in town
    • Better visibility into multi-rig workflows
  • Convoy issue type - Reactive completion tracking

    • Issues that complete when all tracked items complete
    • tracks relation type for convoy membership
  • Agent identity trailers - prepare-commit-msg hook

    • Automatically adds agent identity to commits
    • Structured labels for agent beads
  • Daemon RPC endpoints - Config and mol stale

    • Remote config queries via daemon
    • Stale molecule detection endpoint
  • Non-TTY auto-detection - Cleaner output in pipes

    • Automatically adjusts output for non-interactive use

Fixed

  • Git hook chaining now works correctly (GH#816)
  • .beads/redirect not committed - Prevents worktree conflicts (GH#814)
  • bd sync with sync-branch - Fixes worktree copy direction (GH#810, #812)
  • sync.branch validation - Rejects main/master as sync branch (GH#807)
  • Read operations read-only - No more database writes on list/ready/show (GH#804)
  • bd list defaults - Non-closed issues, 50 limit to protect context (GH#788)
  • External direct-commit bypass - When sync.branch configured (bd-n663)
  • Migration 022 syntax - SQL error on v0.30.3 upgrade path
  • MCP redirect support - Plugin follows .beads/redirect files
  • Jira sync error - Better message when Python script not found (GH#803)
  • Doctor false positives - For molecule/wisp prefix variants
  • BD_ACTOR in direct mode - Consistent actor handling
  • Label accumulation - When updating agent role_type/rig

[0.41.0] - 2025-12-29

Added

  • bd swarm commands - Multi-agent batch work coordination

    • bd swarm create - Create swarm from epic with children
    • bd swarm status - Show swarm progress and blocked work
    • bd swarm validate - Validate swarm structure and dependencies
    • N+1 query fix for blocked checks (performance optimization)
  • bd repair command - Orphaned reference repair

    • Detect and repair orphaned foreign key references
    • Support for comments/events orphan detection
    • --json flag for machine-readable output
    • Transaction safety with backup and dirty_issues marking
  • bd compact --purge-tombstones - Dependency-aware cleanup

    • Remove tombstones while respecting dependency order
    • Safe cleanup that won't break DAG structure
  • bd init --from-jsonl - Manual cleanup preservation

    • Initialize database from curated JSONL file
    • Preserves manual edits made to JSONL
  • bd human command - Focused help menu

    • Human-friendly command reference
    • Quick overview without CLI verbosity
  • bd show --short - Compact output mode

    • Brief issue summary for scripting
    • Less verbose than default format
  • bd delete --reason - Audit trail for deletions

    • Optional reason stored in activity log
    • Better traceability for issue cleanup
  • hooked status - Hook-based work assignment

    • New status for issues assigned to agent hooks
    • Enables autonomous agent work pickup
  • mol_type schema field - Molecule classification

    • Track molecule type (patrol, work, etc.)
    • New migration (adds schema field)
  • Agent ID canonical naming - Validation update

    • Updated validation for orchestrator naming conventions
    • Supports rig/role/name format

Fixed

  • --var flag allows commas in values (GH#786)

    • Variables like --var files=a.go,b.go now work correctly
    • Parser respects quoted values
  • bd sync in bare repo worktrees (GH#785)

    • Fixed sync failure when working in worktrees of bare repos
    • Correctly detects git configuration
  • bd delete --cascade recursive deletion (GH#787)

    • Now properly deletes all transitive dependents
    • Previously only deleted direct children
  • bd doctor pre-push hook detection (GH#799)

    • No longer falsely reports hook issues
    • Correctly identifies bd-managed hooks
  • Gitignore fork protection (GH#796)

    • Removed negations that could override protection
    • Safer fork handling
  • Illumos/Solaris disk space check (GH#798)

    • Added platform support for disk space detection
    • Expands OS compatibility
  • Pre-migration orphan cleanup - Chicken-and-egg fix

    • Clean orphans before migration to avoid failures
    • Smoother upgrade path
  • hq- prefix routing - Town root discovery

    • Correctly finds routes.jsonl from anywhere in town
    • Fixes cross-rig routing for HQ beads
  • Config.yaml database override warning

    • Shows warning when config overrides db location
    • Helps debug unexpected behavior
  • normalizeBeadsRelPath edge case - Similar prefix handling

    • Fixes path normalization for similar prefix names
    • e.g., beads vs beads-sync
  • bd doctor --fix redirect handling

    • Properly follows .beads/redirect files
    • Limited verbose output for cleaner runs

Changed

  • CLI command consolidation - Reduced surface area

    • Grouped related commands under parent commands
    • Cleaner bd --help output
  • Code organization - File size limits

    • Split large cmd/bd files to meet 800-line limit
    • init.go: 1928 → 705 lines
    • Improved maintainability
  • Documentation updates

    • Replace Epic Planning with Ready Front model
    • Add components overview (CLI vs Plugin vs MCP)
    • Add installation method comparison table

Internal

  • Extract IssueDetails to shared type
  • Export FollowRedirect and consolidate implementations
  • Extract shared getEpicChildren helper for swarm commands
  • Extract hashFieldWriter to reduce ComputeContentHash repetition
  • Break up runCook (275 lines) into focused helpers
  • Break up flushToJSONLWithState (280 lines) into focused helpers
  • Extract shared importFromJSONLData function
  • Consolidate duplicated step collection functions
  • Add git helper and guard annotations for tests
  • Fix golangci-lint errors (errcheck, gosec, unparam)

[0.40.0] - 2025-12-28

Added

  • bd worktree command - Parallel development support

    • Manage git worktrees with beads integration
    • Enables working on multiple issues simultaneously
  • bd slot commands - Agent bead slot management

    • Track agent assignments with dedicated slot operations
    • Supports multi-agent orchestration workflows
  • bd agent state command - ZFC-compliant state reporting

    • Report agent state in standardized format
    • Enables machine-readable status checks
  • bd doctor --deep - Full graph integrity validation

    • Deep validation of bead dependency graphs
    • Includes agent bead integrity checks
  • Agent bead support - New bead types

    • type=agent and type=role for agent tracking
    • Agent-specific fields in schema (migration 030)
    • Agent ID pattern validation on create
  • Computed .parent field - JSON output enhancement

    • Parent field included in JSON for convenience
    • Simplifies integration with external tools
  • Auto-bypass daemon for wisp operations - Performance

    • Ephemeral wisp operations skip daemon overhead
    • Faster patrol cycle processing
  • Pour warning for vapor-phase formulas - Safety

    • Warning when pouring formulas designed for wisps
    • Prevents accidental persistent molecule creation

Fixed

  • O(2^n) → O(V+E) cycle detection (GH#775) - Performance

    • Replaced exponential algorithm with linear DFS
    • Dramatic speedup for large dependency graphs
  • Import hash mismatch warnings - Data integrity

    • Update jsonl_file_hash on import operations
    • Prevents spurious hash mismatch warnings

Deprecated

The following commands are deprecated and will be removed in v1.0.0:

  • bd relate → use bd dep relate instead
  • bd unrelate → use bd dep unrelate instead
  • bd daemons → use bd daemon <subcommand> instead
  • bd cleanup → use bd admin cleanup instead
  • bd compact → use bd admin compact instead
  • bd reset → use bd admin reset instead
  • bd comment → use bd comments add instead
  • bd template → use bd mol instead
  • bd templates → use bd formula list instead
  • bd template show → use bd mol show instead
  • bd template bond → use bd mol bond instead
  • bd detect-pollution → use bd doctor --check=pollution instead
  • bd migrate-hash-ids → use bd migrate hash-ids instead
  • bd migrate-tombstones → use bd migrate tombstones instead
  • bd migrate-sync → use bd migrate sync instead
  • bd migrate-issues → use bd migrate issues instead

All deprecated commands continue to work but print a warning. Update your scripts and muscle memory before v1.0.0 to avoid breakage.

Changed

  • Community tools documentation (GH#772, GH#776)

    • Consolidated into dedicated docs page
    • Added opencode-beads to community tools list
  • Activity feed improvements - Better context

    • Shows title and assignee in activity entries
    • More informative bd activity output

[0.39.1] - 2025-12-27

Added

  • bd where command - Show active beads location

    • Displays the resolved database path after following redirects
    • Helpful for debugging multi-rig setups
  • --parent flag for bd update - Reparent issues

    • Move issues between epics with bd update <id> --parent=<new-parent>
    • Supports clearing parent with --parent=none
  • Redirect info in bd prime - Multi-clone support

    • Shows when database is redirected to another location
    • Improves visibility into routing behavior

Fixed

  • Doctor follows redirects - Multi-clone compatibility

    • bd doctor now correctly follows database redirects
    • Prevents false negatives when running from rig roots
  • Remove 8-char prefix limit (GH#770) - bd rename-prefix

    • Removed arbitrary length restriction on issue prefixes
    • Allows longer, more descriptive prefixes

Changed

  • Git context consolidation - Internal refactor
    • Unified git context into single cached struct
    • Reduces redundant git operations

Documentation

  • Database Redirects section - ADVANCED.md

    • Comprehensive documentation for redirect feature
    • Explains multi-clone integration patterns
  • Community Tools update (GH#771) - README.md

    • Added opencode-beads to community tools list

[0.39.0] - 2025-12-27

Added

  • bd orphans command (GH#767) - Detect orphaned issues

    • Finds issues mentioned in git commits that were never closed
    • DRY refactoring of orphan detection from bd doctor
    • Helps maintain issue hygiene in larger projects
  • bd admin parent command - Consolidated admin tools

    • bd admin cleanup - Clean up unused data
    • bd admin compact - Compact the database
    • bd admin reset - Reset to initial state
    • Reduces top-level command clutter
  • --prefix flag for bd create - Cross-rig issue creation

    • Create issues in other rigs from any directory
    • Accepts prefix (bd-), short form (bd), or rig name (beads)
    • Works with existing routes.jsonl routing

Changed

  • bd mol catalog → bd formula list - Command rename

    • Aligns with formula-based terminology
    • bd formula list shows available molecule templates
  • bd info --thanks - Relocated thanks command

    • Contributors list moved under bd info
    • Reduces command namespace pollution
  • Removed unused commands

    • bd pin, bd unpin, bd hook removed
    • Functionality covered by orchestrator molecule commands
    • Cleaner separation between beads (data) and orchestration
  • bd doctor --check=pollution - Integrated test pollution check

    • Detects test artifacts left in production database
    • Previously standalone bd detect-pollution command

Fixed

  • macOS codesigning - Fixed bump-version.sh --install

    • Ad-hoc signing for local installations
    • Prevents macOS from quarantining the binary
  • Lint errors and Nix vendorHash (GH#769)

    • Resolved golangci-lint issues
    • Updated Nix package hash for reproducible builds

Documentation

  • Issue Statuses section - CLI_REFERENCE.md

    • Comprehensive status lifecycle documentation
    • Clear explanations of open, pinned, in_progress, blocked, deferred, closed, tombstone
  • Consolidated UI_PHILOSOPHY files (GH#745)

    • Merged duplicate philosophy docs
    • Single source of truth for UI design principles
  • README and PLUGIN.md fixes (GH#763)

    • Corrected installation instructions
    • Updated plugin documentation

[0.38.0] - 2025-12-27

Added

  • Prefix-based routing - Cross-rig command routing

    • bd commands auto-route to correct rig based on issue ID prefix
    • Routes stored in ~/gt/.beads/routes.jsonl
    • Enables seamless multi-rig workflows from any directory
  • Cross-rig ID auto-resolve - Smarter dependency handling

    • bd dep add auto-resolves issue IDs across different rigs
    • No need to specify full paths for cross-rig dependencies
  • bd mol pour/wisp subcommands - Reorganized command hierarchy

    • bd mol pour for persistent molecules
    • bd mol wisp for ephemeral workflows
    • Cleaner organization under bd mol namespace
  • Comments display in bd show (GH#177) - Enhanced issue details

    • Comments now visible in issue output
    • Shows full discussion thread for issues
  • created_by field on issues (GH#748) - Creator tracking

    • Track who created each issue for audit trail
    • Useful for multi-agent workflows
  • Database corruption recovery (GH#753) - Robust doctor repairs

    • bd doctor --fix can now auto-repair corrupted SQLite databases
    • JSONL integrity checks detect and fix malformed entries
    • Git hygiene checks for stale branches
  • Chaos testing for releases - Thorough validation

    • --run-chaos-tests flag in release script
    • Exercises edge cases and failure modes
  • Pre-commit config - Local lint enforcement

    • Consistent code quality before commits

Changed

  • CLI consolidation - Reduced top-level command clutter

    • bd cleanup, bd compact, bd reset → bd admin cleanup|compact|reset
    • bd migrate-* → bd migrate hash-ids|issues|sync|tombstones
    • bd cook → bd formula cook
    • bd detect-pollution → bd doctor --check=pollution
    • bd quickstart hidden, use docs/QUICKSTART.md instead
    • Hidden aliases maintain backwards compatibility with deprecation notices
  • Sync backoff and tips consolidation (GH#753) - Smarter daemon

    • Daemon uses exponential backoff for sync retries
    • Tips consolidated from multiple sources
  • Wisp/Ephemeral naming finalized - wisp is canonical

    • bd mol wisp is the correct command
    • Internal API uses "ephemeral" but CLI uses "wisp"

Fixed

  • Comments display position (GH#756) - Formatting fix

    • Comments now display outside dependents block
    • Proper visual hierarchy in bd show output
  • no-db mode storeActive (GH#761) - JSONL-only fix

    • storeActive correctly set in no-database mode
    • Fixes issues with JSONL-only installations
  • --resolution alias (GH#746) - Backwards compatibility

    • Restored --resolution as alias for --reason on bd close
  • bd graph with daemon (GH#751) - Daemon compatibility

    • Graph generation works when daemon is running
    • No more conflicts between graph and daemon operations
  • created_by in RPC path (GH#754) - Daemon propagation

    • Creator field correctly passed through daemon RPC
  • Migration 028 idempotency (GH#757) - Safe re-runs

    • Migration handles partial or repeated runs gracefully
    • Checks for existing columns before adding
  • Routed ID daemon bypass - Cross-rig show

    • bd show with routed IDs bypasses daemon correctly
    • Storage connections closed per iteration to prevent leaks
  • Modern git init (GH#753) - Test compatibility

    • Tests use --initial-branch=main for modern git versions
  • golangci-lint clean (GH#753) - All platforms

    • Resolved all lint errors across platforms

Improved

  • Test coverage - Comprehensive testing
    • Doctor, daemon, storage, and RPC client paths covered
    • Chaos testing integration for edge cases

[0.37.0] - 2025-12-26

Added

  • Gate commands - Async coordination for agent workflows

    • bd gate create - Create gates with await conditions (gh:run, gh:pr, timer, human, mail)
    • bd gate eval - Evaluate timer and GitHub gates (checks run completion, PR merge)
    • bd gate approve - Approve human gates
    • bd gate show/list/close/wait - Full gate lifecycle management
  • bd close --suggest-next (GH#679) - Smart workflow continuation

    • Shows newly unblocked issues after closing
    • Helps agents find next actionable work
  • bd ready/blocked --parent (GH#743) - Epic-scoped filtering

    • Filter issues by parent bead or epic
    • Enables focused work within a subtree
  • TOML support for formulas - Alternative format

    • .formula.toml files alongside JSON support
    • Human-friendly authoring option
  • Fork repo auto-detection (GH#742) - Contributor workflow

    • Detects fork repositories during init
    • Offers to configure .git/info/exclude for stealth mode
  • Control flow operators - Advanced formula composition

    • loop operator for iterating over collections
    • gate operator for conditional execution
    • Condition evaluator for gates and loops
    • Control flow integrates with aspect composition
  • Aspect composition - Cross-cutting workflow concerns

    • aspects field in formulas for reusable patterns
    • Aspects match steps by type, title pattern, or custom criteria
    • Enables logging, retries, notifications as composable concerns
  • Runtime expansion types - Dynamic step generation

    • on_complete - Generate steps when a step completes
    • for-each - Expand steps for each item in a collection
    • Enables dynamic workflows based on runtime state
  • bd formula list/show - Formula discovery commands

    • bd formula list - List available formulas in search paths
    • bd formula show <name> - Display formula definition and metadata
  • bd mol stale - Detect complete-but-unclosed molecules

    • Finds molecules where all children are closed but molecule is open
    • Helps identify forgotten cleanup tasks
  • Stale molecules check in doctor - Proactive detection

    • bd doctor now warns about stale molecules
    • Integrated with bd mol stale detection
  • Distinct ID prefixes - Clear entity type identification

    • Protos: bd-proto-xxx prefix
    • Molecules: bd-mol-xxx prefix
    • Wisps: bd-wisp-xxx prefix
    • Makes entity type immediately visible in IDs
  • no-git-ops config (GH#593) - Manual git control

    • bd config set no-git-ops true to disable auto git operations
    • bd prime outputs stealth-mode protocol when enabled
    • Useful for custom git workflows or manual commit review

Changed

  • Formula format: YAML → JSON - Standardized format

    • Formulas now use .formula.json extension
    • JSON provides better tooling support and validation
    • Existing YAML formulas need migration
  • Removed bd mol run - Orchestration delegated to orchestrator

    • Molecule execution now handled by orchestrator commands
    • bd focuses on issue tracking primitives
    • Use orchestrator's molecule runner for execution
  • Simplified wisp architecture - Single database model

    • Wisps stored in main database with Wisp=true flag
    • Removed separate .beads-wisp/ directory
    • Wisps filtered from JSONL export automatically

Fixed

  • Gate await fields preserved during upsert - Multirepo sync

    • Gate fields no longer cleared during multi-repo sync operations
    • Enables reliable gate coordination across clones
  • Tombstones retain closed_at timestamp - Soft delete metadata

    • Tombstones preserve the original close time
    • Maintains accurate timeline for deleted issues
  • Git detection caching - Performance improvement

    • Cache git worktree/repo detection results
    • Eliminates repeated slowness on worktree checks
  • Windows MCP graceful fallback (GH#387) - Platform compatibility

    • Daemon mode falls back gracefully on Windows
    • Prevents crashes when daemon unavailable
  • Windows npm postinstall file locking (GH#670) - Install reliability

    • Handles file lock errors during npm install
    • Improves installation success on Windows
  • installed_plugins.json v2 format (GH#741) - Claude Code compatibility

    • bd doctor now handles both v1 and v2 plugin file formats
    • Fixes "cannot unmarshal array" error on newer Claude Code versions
  • git.IsWorktree() hang on Windows (GH#727) - Init safety

    • Added isGitRepo() guard before calling git.IsWorktree()
    • Fixes bd init hanging outside git repositories on Windows
  • Skill files deleted by bd sync (GH#738) - Sync safety

    • bd sync no longer deletes skill files in .claude/
    • Prevents accidental removal of Claude Code configuration
  • doctor check false positives (GH#709) - Cleaner diagnostics

    • Skips interactions.jsonl and molecules.jsonl in sync checks
    • These files are runtime state, not sync targets
  • FatalErrorRespectJSON - Consistent error output

    • All commands respect --json flag for error output
    • Errors return proper JSON structure when flag is set
  • bd sync commits non-.beads files - Sync isolation

    • Sync operations now only commit .beads/ directory changes
    • Prevents unintended commits of working directory files
  • Content-level merge for divergence - Better conflict resolution

    • Divergence recovery uses content-level merging
    • Reduces false conflicts during multi-clone sync
  • Child→parent dep fix opt-in - Migration control

    • --fix-child-parent flag required for automatic fix
    • Prevents unexpected dependency modifications
  • Aspect self-matching recursion - Formula safety

    • Aspects cannot match themselves during expansion
    • Prevents infinite recursion in aspect composition
  • Map expansion nested matching - Correct child handling

    • Map expansion now correctly matches nested child steps
    • Fixes missing expansions in hierarchical formulas
  • Mol/wisp ID prefix combination - Correct ID generation

    • Database prefix combined with type prefix correctly
    • IDs like bd-mol-xxx instead of bdmol-xxx
  • mol run hierarchical children (bd-c8d5, bd-drcx) - Complete loading

    • bd mol run loads all hierarchical children
    • Supports title lookup for molecule identification
  • Wisps excluded from sync (bd-687g, bd-9avq) - Export correctness

    • Wisps filtered from JSONL export and nodb operations
    • Ephemeral molecules don't pollute shared state

Documentation

  • MOLECULES.md rewrite - Workflow-first structure

    • Reorganized around practical workflows
    • Clearer distinction between protos, molecules, wisps
  • Molecular chemistry docs - Conceptual foundation

    • Phase metaphor: solid (proto) → liquid (mol) → vapor (wisp)
    • Pour, bond, distill, squash operations explained

[0.36.0] - 2025-12-24

Added

  • Formula system (bd-weu8, bd-wa2l) - Declarative workflow templates

    • bd cook <formula> - Execute a formula template with variable interpolation
    • Formula files (.formula.json) support inheritance via extends
    • needs and waits_for fields for dependency declarations
    • --prefix flag for custom issue prefix when cooking
    • Search paths: .beads/formulas/, ~/.beads/formulas/, ~/gt/.beads/formulas/
  • Gate issue type - Async coordination primitives

    • bd gate create <name> - Create a gate for coordinating parallel work
    • bd gate open <id> - Open a gate (unblock waiters)
    • bd gate close <id> - Close a gate (block new work)
    • Gates integrate with waits-for dependencies for fanout patterns
  • bd list viewer enhancements (#729) - Built-in terminal UI

    • --pretty - Colorized, formatted output for human viewing
    • --watch - Live-updating view (refreshes on changes)
  • bd search filters - Enhanced search capabilities

    • --after, --before - Date range filtering
    • --priority - Exact priority match
    • --content - Full-text content search
  • bd compact --prune - Standalone tombstone pruning

    • Prune old tombstones without full compaction
    • Configurable retention period
  • bd export --priority - Exact priority filter for exports

  • --resolution alias (GH#721) - Alternative to --reason on bd close

    • bd close bd-42 --resolution "Fixed in commit abc123"
  • Database size check in doctor (#724) - Issue count threshold warnings

    • Warns when issue count exceeds configurable threshold
    • Helps identify when cleanup or compaction is needed
  • Config override notifications (#731) - Transparency for config sources

    • Shows when environment variables override config file values
    • Helps debug unexpected behavior from env overrides
  • Windows code signing infrastructure - Signed Windows binaries

    • Code signing certificate integration for Windows releases
    • Improved trust and installation experience on Windows
  • RPC endpoints for monitoring (bd-0oqz, bd-l13p)

    • GetMoleculeProgress - Query molecule execution state
    • GetWorkerStatus - Query worker health and current task
  • Cross-database molecule spawning

    • bd mol run can spawn in external beads databases
    • Enables cross-project workflow orchestration
  • Config-based close hooks - Custom scripts on issue close

    • Configure hooks in .beads/config.yaml
    • Run validation or notification scripts when issues close

Changed

  • Removed bd mol spawn - Use pour/wisp only

    • bd pour <proto> for persistent molecules
    • bd wisp create <proto> for ephemeral molecules
    • Simplifies mental model: pour = liquid (persistent), wisp = vapor (ephemeral)
  • bd ready excludes workflow types - Cleaner ready queue

    • Gates and other workflow coordination types excluded by default
    • Use --include-workflow to see all types
  • Natural language activation (#718) - Improved Claude integration

    • Enhanced activation patterns for natural language commands
    • Anthropic 2025 API compliance updates

Fixed

  • Dots in prefix handling (GH#664) - SQLite extractParentChain fix

    • Prefixes containing dots (e.g., my.project) now work correctly
    • Parent chain extraction uses proper escaping
  • allowed_prefixes config respected - Import validation

    • Import now respects allowed_prefixes configuration
    • Prevents importing issues with unauthorized prefixes
  • Child counter updates (GH#728) - Explicit child ID creation

    • child_counters table updated when explicit child IDs created
    • Prevents ID collision when mixing auto and explicit child IDs
  • Comment timestamps preserved (#735) - Import fidelity

    • created_at timestamps on comments preserved during import
    • Maintains audit trail for imported issues
  • sync.remote config respected (#736) - Daemon sync operations

    • Daemon respects configured remote for sync operations
    • Fixes sync to wrong remote when multiple remotes configured
  • Export requires -o flag (#733) - Explicit output control

    • bd export now requires -o flag to write to file
    • Prevents accidental stdout pollution in scripts
  • YAML config key normalization (#732) - Consistent config parsing

    • Config keys normalized to canonical format (snake_case)
    • sync.remote and sync_remote both work
  • JSON output standardization - Consistent API

    • Empty arrays return [] not null
    • Error responses have consistent structure
    • All commands with --json follow same patterns
  • MCP Claude Code compatibility - Tool schema fix

    • Added output_schema=None to MCP tools
    • Fixes "Invalid schema" errors in Claude Code
  • Windows file locking - Better Windows support

    • Proper file handle cleanup prevents locking issues
    • Fixes "file in use" errors on Windows
  • Template commands with daemon - Daemon mode compatibility

    • bd pour, bd wisp create work correctly with daemon running
  • Startup config to config.yaml (GH#536) - Config persistence

    • Startup wizard writes to config.yaml, not SQLite
    • Configuration survives database recreation
  • Multi-hyphen prefixes (GH#422) - Prefix parsing

    • Prefixes like my-project-name parsed correctly
    • Fixes issue ID extraction with complex prefixes
  • --pour flag in bond operations - Phase control

    • bd mol bond --pour correctly forces liquid phase
  • Stealth mode gitignore (GH#704) - Local-only exclusion

    • Uses .git/info/exclude instead of global gitignore
    • Stealth mode truly local to the repository
  • Pinned field import - Field preservation

    • pinned field preserved during JSONL import
  • External deps in orphan check - Migration fix

    • External dependencies excluded from orphan validation
    • Fixes spurious migration warnings
  • Child→parent dependency detection - Prevents LLM temporal reasoning trap

    • Epic children can no longer depend on their parent epic
    • Blocks a common AI mistake: LLMs use temporal reasoning for "phases" and invert dependencies
    • Example: "Phase 1 before Phase 2" triggers "Phase 1 blocks Phase 2" → WRONG
    • Correct model: "Phase 2 needs Phase 1" → bd dep add phase2 phase1
    • Clear error message explains the anti-pattern when detected

Improved

  • Test coverage - Significant improvements across codebase

    • daemon: 27% → 72%
    • compact: 17% → 82%
    • setup: 28% → 54%
    • storage: interface conformance tests
    • RPC: comprehensive delete handler tests
  • Structured logging - Better observability

    • Daemon uses slog for structured logging
    • Consistent log format across components
  • Code organization - Modular refactoring

    • Split sync.go into focused modules
    • Split queries.go into focused modules
    • Typed JSON response structs (no more map[string]interface{})
  • JSONL size reduction - Smaller exports

    • omitempty on all JSONL fields
    • Removes null/empty values from exports

[0.35.0] - 2025-12-23

Added

  • Dynamic molecule bonding - Attach templates to running molecules at runtime

    • bd mol bond <proto> <target> --ref <issue-id> - Inject template steps dynamically
    • Enables responsive workflows: attach error recovery, expand scope mid-execution
    • Templates can reference existing issue data via --ref for variable interpolation
  • bd activity command - Real-time state feed for workflow monitoring

    • Stream mutation events as they happen (JSON format)
    • Monitor molecule progress without polling
    • Useful for dashboards, debugging, and automation
  • waits-for dependency type - Fanout coordination gates

    • bd dep add <gate> <step> --type waits-for - Gate waits for all steps
    • Unlike blocks, waits-for is used for parallel coordination (fan-in pattern)
    • Gate unblocks only when ALL waits-for dependencies complete
  • Parallel step detection - Auto-identify parallelizable work

    • Molecules automatically detect steps that can run concurrently
    • bd mol show displays parallel-capable step groups
    • Helps agents optimize execution order
  • Molecule navigation commands (bd-sal9, bd-ieyy) - Track progress through workflows

    • bd mol current - Show current step and molecule progress
    • bd close --continue - Auto-advance to next step in molecule
    • bd close --no-auto - Close without auto-claiming next step
  • bd list --parent flag - Filter issues by parent

    • bd list --parent=bd-xyz --status=open shows open children of bd-xyz
    • Useful for epic progress tracking and molecule step listing
  • Conditional bond type - Run steps only when predecessors fail

    • bd mol bond --type conditional-blocks for error-handling workflows
    • Step B runs only if step A fails (useful for fallback paths)
  • External dependency display - Show cross-project deps in tree view

    • bd dep tree now displays external dependencies with repo prefix
    • Satisfied external deps filtered from blocked issues
  • Performance optimization indexes (bd-bha9, bd-a9y3) - Faster queries for large databases

    • Added indexes for common query patterns
    • Improved GetReadyWork and SearchIssues performance

Changed

  • Consolidated doctor --fix (GH#715, bd-bqcc) - Single command for all fixes

    • bd doctor --fix now handles: hooks, permissions, sync branch, schema
    • Removed individual bd migrate, bd hooks install suggestions
    • Daemon health check integrated into doctor flow
  • Auto_pull config (GH#707) - Periodic remote sync in event-driven mode

    • bd config set daemon.auto_pull true enables pull every 5 minutes
    • Prevents staleness when git changes happen outside daemon's watch

Fixed

  • Rich mutation events for status changes - Events now include full context

    • Status transitions emit complete before/after state
    • Enables better monitoring and auditing
  • External deps filtered from GetBlockedIssues - Correct blocking calculation

    • External dependencies no longer incorrectly block local work
    • Only local blocking deps affect ready queue
  • Parallel execution migration race (GH#720) - Multiple daemons no longer corrupt DB

    • Added file-based migration lock to prevent concurrent schema changes
    • Retry logic when migration lock is held by another process
  • bd create -f with daemon (GH#719) - File flag now works in daemon mode

    • -f/--file flag properly forwarded through RPC
  • FindJSONLInDir interactions.jsonl (GH#709) - No longer picks wrong file

    • Skip interactions.jsonl when discovering JSONL files
  • Daemon duplicate log messages (PR#713) - Reduced log spam

  • Worktree health check redundancy (PR#711) - Integrated into CreateBeadsWorktree

  • Diverged sync branch handling (GH#697) - Auto-recovery on push failure

  • Tombstones in JSONL export (GH#696) - Deletions now propagate correctly

Refactored

  • Remove legacy autoflush code paths
  • Consolidate duplicate path-finding utilities
  • Replace map[string]interface{} with typed JSON structs
  • Migrate sort.Slice to slices.SortFunc
  • Add testEnv helpers, reduce SQLite test file sizes

[0.34.0] - 2025-12-22

Added

  • Wisp commands - Full ephemeral molecule management

    • bd wisp create <proto> - Instantiate proto as ephemeral wisp (solid→vapor)
    • bd wisp list - List all wisps with stale detection
    • bd wisp gc - Garbage collect orphaned wisps
    • Wisps have Wisp=true flag and are not exported to JSONL (never sync)
  • Chemistry UX commands - Phase-aware molecule operations

    • bd pour <proto> - Instantiate proto as persistent mol (solid→liquid)
    • bd mol bond --wisp - Force spawn as vapor when attaching to mol
    • bd mol bond --pour - Force spawn as liquid when attaching to wisp
    • Squash clears Wisp flag, promoting to persistent (exported to JSONL)
  • Cross-project dependencies (bd-66w1, bd-om4a) - Reference issues across repos

    • external:<repo>:<id> dependency syntax
    • bd ship <id> --to <repo> - Ship issues to other beads repos
    • bd ready filters by external dependency satisfaction
    • Configure additional repos in .beads/config.yaml
  • Orphan detection in bd doctor - Find issues with missing parents

    • Detects parent-child relationships pointing to deleted issues
    • Suggests fix commands for orphaned issues

Changed

  • Multi-repo config uses YAML (GH#683) - bd repo add/remove now writes to .beads/config.yaml
    • Fixes disconnect where CLI wrote to DB but hydration read from YAML
    • bd repo remove now cleans up hydrated issues from removed repo
    • Breaking: bd repo add no longer accepts optional alias argument

Fixed

  • Wisp flag handling - Wisps now use Wisp=true flag in main database (not exported to JSONL)
  • Prefix validation in multi-repo mode (GH#686) - Skip validation for external repos
  • Empty config values (GH#680, GH#684) - Handle gracefully in getRepoConfig()
  • Doctor UX improvements (GH#687) - Better diagnostics and daemon integration
  • Orphaned test file - Removed repo_test.go with undefined functions

[0.33.2] - 2025-12-21

[0.33.1] - 2025-12-21

Changed

  • Ephemeral → Wisp rename - Aligns with Steam Engine metaphor (wisps = ephemeral vapor)
    • JSON field changed from ephemeral to wisp (breaking change for API consumers)
    • CLI flag changed from --ephemeral to --wisp on bd cleanup
    • SQLite column remains ephemeral (no migration needed)

Fixed

  • Lint error in mail.go - Added nosec directive for mail delegate exec.Command

[0.33.0] - 2025-12-21

Added

  • Wisp molecules - Support for ephemeral wisps
    • bd wisp create creates wisp issues that live only in SQLite
    • Wisp issues never export to JSONL (prevents zombie resurrection)
    • Use bd pour for persistent mols, bd wisp create for ephemeral wisps
    • bd mol squash compresses wisp children into a digest issue
    • --summary flag allows agents to provide AI-generated summaries

Fixed

  • Comments not deleted with issue - DeleteIssue now cascades to comments table

[0.32.1] - 2025-12-21

Added

  • MCP output control parameters (PR#667) - Reduce context window usage by up to 97%

    • brief - Return minimal responses: BriefIssue for reads, OperationResult for writes
    • brief_deps - Full issue with compact dependencies
    • fields - Custom field projection with validation
    • max_description_length - Truncate long descriptions
    • New models: BriefIssue, BriefDep, OperationResult
    • Default brief=True for writes (minimal confirmations)
  • MCP filtering parameters - Align MCP tools with CLI capabilities

    • labels / labels_any - AND/OR label filtering
    • query - Title search (case-insensitive)
    • unassigned - Filter to unassigned issues
    • sort_policy - Sort ready work by hybrid/priority/oldest

Fixed

  • Pin field not in allowed update fields
    • bd update --pinned was failing with "invalid field" error
    • Added pinned to allowedUpdateFields and importer

[0.32.0] - 2025-12-20

Changed

  • Removed bd mail commands - Mail is orchestration, not data plane
    • Removed: bd mail send, bd mail inbox, bd mail read, bd mail ack, bd mail reply
    • The underlying data model is unchanged: type=message, Sender, Ephemeral, replies_to fields remain
    • Orchestration tools should implement mail interfaces on top of the beads data model
    • This follows the principle that beads is a data store, not a workflow engine

Fixed

  • Symlink preservation in atomicWriteFile (PR#665) - Thanks @qmx

    • bd setup claude no longer clobbers nix/home-manager managed ~/.claude/settings.json
    • New ResolveForWrite() helper writes to symlink target instead of replacing symlink
  • Broken link in examples (GH#666) - Fixed LABELS.md link in multiple-personas example

[0.31.0] - 2025-12-20

Added

  • bd defer / bd undefer commands - New "deferred" status for icebox issues

    • Issues that are deliberately postponed, not blocked by dependencies
    • Deferred issues excluded from bd ready and shown with ❄️ snowflake styling
    • Full support in MCP server, graph views, and statistics
  • Agent audit trail (GH#649) - Append-only logging for AI agent interactions

    • New .beads/interactions.jsonl audit file
    • bd audit record - Log LLM calls, tool calls, or pipe JSON via stdin
    • bd audit label <id> - Append quality labels (good/bad) for dataset curation
    • bd compact --audit - Optionally log compaction prompts/responses
    • Audit entries are immutable; labels create new referencing entries
  • Directory-aware label scoping (GH#541) - Auto-filter issues by directory in monorepos

    • Configure directory.labels to map paths to label filters
    • bd ready and bd list auto-apply when in matching directories
    • Example: packages/maverick: maverick shows only maverick-labeled issues
  • Molecules catalog - Separate storage for template molecules

    • Templates now live in molecules.jsonl, distinct from work items
    • Hierarchical loading: built-in → town → user → project
    • Molecules use mol-* ID namespace with is_template: true
  • Windows winget manifest (GH#524) - Prepare beads for Windows Package Manager

    • Added manifest files for winget submission
    • Once merged to microsoft/winget-pkgs: winget install SteveYegge.beads
  • Git commit configuration (GH#600) - Control beads auto-commit behavior

    • git.author - Override commit author (useful for bots)
    • git.no-gpg-sign - Disable GPG signing (fixes Touch ID prompts)
  • Require description config (GH#596) - Enforce descriptions on issue creation

    • Set create.require-description: true to error on missing descriptions
    • Also supports BD_CREATE_REQUIRE_DESCRIPTION env var

Changed

  • bd stats merged into bd status (GH#644) - Consolidated status commands

    • stats now an alias for status
    • Colorized output with emoji header
    • Shows all statistics (tombstones, pinned, epics, lead time)
    • Added --no-activity flag to skip git activity parsing
  • Thin hook shims (GH#615) - Hooks now delegate to bd hooks run

    • Eliminates hook version drift after upgrades
    • No more manual bd hooks install --force needed
    • Shims use # bd-shim v1 marker (format version, not bd version)
  • MCP context tool consolidation - Merged 3 tools into 1

    • Combined set_context, where_am_i, init into single context tool
    • Actions: set (default with workspace_root), show (default), init
  • Doctor improvements (GH#656) - Enhanced diagnostics and testing

    • Visual improvements with grouped output
    • Comprehensive test coverage added
    • Fixed gosec warnings
    • Contributed by @rsnodgrass

Fixed

  • relates-to cycle detection (GH#661) - Exclude relates-to from cycle detection

    • Relates-to links are bidirectional by design, not cycles
  • Doctor .local_version check (GH#662) - Check correct version file

    • Now checks .local_version instead of deprecated LastBdVersion
  • Doctor Claude plugin link (GH#623) - Updated broken documentation link

  • Read-only gitignore in stealth mode (GH#663) - Print manual instructions instead of failing

    • When global gitignore is read-only (e.g., symlink to immutable location), shows what to add manually
    • Contributed by @qmx

[0.30.7] - 2025-12-19

Fixed

  • bd graph nil pointer crash (#657) - Fixed crash when running bd graph on epics

    • renderGraph() was passed nil instead of the subgraph, causing panic in computeDependencyCounts()
  • Windows npm installer file lock (#652) - Fixed installation failure on Windows

    • The download file handle wasn't fully closed before extraction attempted
    • Now properly waits for file.close() callback before proceeding

[0.30.6] - 2025-12-18

Added

  • bd graph dependency counts - Graph command shows dependency counts using subgraph formatting
  • types.StatusPinned - New status for persistent beads that survive cleanup

Fixed

  • CRITICAL: Dependency resurrection bug - Fixed 3-way merge to respect dependency removals
    • mergeDependencies was using union (additive-only) instead of proper 3-way merge
    • Now removals are authoritative: if either left or right removes a dep, it stays removed
    • This prevented orphaned parent-child relationships from being permanently removed

[0.30.5] - 2025-12-18

Removed

  • YAML simple template system - Removed --from-template flag from bd create
    • Deleted embedded templates: bug.yaml, epic.yaml, feature.yaml
    • Templates are now purely Beads-based (epics with template label)
    • Use bd template instantiate <id> for template workflows

[0.30.4] - 2025-12-18

Added

  • bd template instantiate - Create beads issues from YAML workflow templates
    • bd template instantiate <file.yaml> - Create issues from workflow definitions
    • --assignee <identity> flag for auto-assignment during instantiation
    • Supports multi-issue workflows with dependency chains
    • Templates define issue properties (title, type, priority, dependencies)

Changed

  • bd mail inbox --identity - Fixed to properly filter by identity parameter

Fixed

  • Orphan detection warnings - No longer warns about closed issues or tombstones
    • Previously bd doctor reported false positives for completed dependencies

Removed

  • Legacy MCP Agent Mail integration - Removed obsolete mcp_agents package
  • YAML workflow execution system - Replaced by simpler template instantiation

Notes

  • Experimental edges: The new graph link fields (relates_to, replies_to, duplicate_of, superseded_by) and mail commands are experimental and subject to change in upcoming releases. Early adopters should expect breaking changes to these APIs.

[0.30.3] - 2025-12-17

Fixed

  • Data loss race condition - Removed unsafe ClearDirtyIssues() method
    • Old method cleared ALL dirty issues, risking data loss if export failed partway
    • All code now uses ClearDirtyIssuesByID() which only clears exported issues
    • Affects: internal/storage/sqlite/dirty.go, internal/storage/memory/memory.go

Closed (Already Implemented)

  • Stale database warning - Commands now warn when database is out of sync with JSONL
  • Staleness check error handling (bd-n4td, bd-o4qy) - Proper warnings and error returns

[0.30.2] - 2025-12-16

Added

  • bd setup droid (GH#598) - Factory.ai (Droid) IDE support

    • Configure beads for use with Factory.ai's Droid
    • Contributed by @jordanhubbard
  • Messaging schema fields - Foundation for inter-agent messaging

    • New message issue type for agent-to-agent communication
    • New fields: sender, wisp, replies_to, relates_to, duplicate_of, superseded_by
    • New dependency types: replies-to, relates-to, duplicates, supersedes
    • Schema migration 019 (automatic on first use)
  • bd mail commands - Inter-agent messaging

    • bd mail send <recipient> -s <subject> -m <body> - Send messages
    • bd mail inbox - List open messages for your identity
    • bd mail read <id> - Display message content
    • bd mail ack <id> - Acknowledge (close) messages
    • bd mail reply <id> -m <body> - Reply to messages (creates threads)
    • Identity via BEADS_IDENTITY env var or .beads/config.json
  • Graph link commands - Knowledge graph relationships

    • bd relate <id1> <id2> - Create bidirectional "see also" links
    • bd unrelate <id1> <id2> - Remove relates_to links
    • bd duplicate <id> --of <canonical> - Mark issue as duplicate (closes it)
    • bd supersede <old> --with <new> - Mark issue as superseded (closes it)
    • bd show --thread - View message threads via replies_to chain
  • Hooks system - Extensible event notifications

    • .beads/hooks/on_create - Runs after issue creation
    • .beads/hooks/on_update - Runs after issue update
    • .beads/hooks/on_close - Runs after issue close
    • .beads/hooks/on_message - Runs after message send
    • Hooks receive issue ID, event type as args, full JSON on stdin
  • bd cleanup --wisp flag - Clean up transient messages

    • Deletes only closed issues with wisp=true
    • Useful for cleaning up messages after swarms complete

Fixed

  • Windows build errors (GH#585) - Fixed gosec lint warnings

    • Contributed by @deblasis
  • Issue ID prefix extraction - Word-like suffixes (e.g., my-project-audit) now parse correctly

    • Previously could incorrectly split on word boundaries

Removed

  • Legacy deletions.jsonl code - Fully migrated to inline tombstones
    • Removed deletions.jsonl from git tracking
    • All deletion tracking now via inline tombstones in issues.jsonl

Documentation

  • Messaging documentation - New docs for messaging system

    • docs/messaging.md - Full messaging reference with examples
    • docs/graph-links.md - Graph link types and use cases
    • Updated AGENTS.md with inter-agent messaging section
  • Windows installation command in upgrade instructions (GH#589)

    • Contributed by @alexx-ftw
  • Aligned bd prime guidance with skill's hybrid TodoWrite approach

[0.30.1] - 2025-12-16

Added

  • bd reset command (GH#505) - Complete beads removal from a repository

    • Removes .beads/ directory and all associated data
    • Cleans up git hooks installed by beads
    • Use --force to skip confirmation prompt
  • bd cleanup --hard flag - Bypass tombstone TTL safety

    • Immediately removes tombstones regardless of age
    • Use when you need to force-clean deleted issues
  • bd update --type flag (GH#522) - Change issue type after creation

    • Convert between task, bug, feature, epic types
    • Example: bd update bd-xyz --type epic
  • bd q silent quick-capture mode (GH#540)

    • Capture issues silently without output
    • Ideal for scripting and automation
  • bd sync --check flag - Pre-sync integrity checks

    • Validates database state before syncing
    • Catches potential issues early
  • bd show displays dependent issue status (#583)

    • Shows status for all blocked-by and blocking issues
    • Better visibility into dependency chains
  • bd daemon --status shows config (#569)

    • Displays daemon configuration in status output
    • Contributed by @crcatala
  • bd daemon --stop-all - Kill all daemon processes

    • Useful for cleanup when multiple daemons are running
  • Auto-disable daemon in git worktrees (#567)

    • Daemon automatically disabled in worktrees for safety
    • Prevents database conflicts between worktrees
  • claude.local.md support - Local-only documentation

    • Add project notes that won't be committed
    • Gitignored by default
  • Auto-add session completion instructions to AGENTS.md

    • New projects get session-close protocol guidance
  • Inline tombstones for soft-delete

    • Deleted issues become tombstones with status: "tombstone" in issues.jsonl
    • Full audit trail: deleted_at, deleted_by, delete_reason, original_type
    • TTL-based expiration (default 30 days) with automatic pruning via bd compact
    • Proper 3-way merge support: fresh tombstones win, expired tombstones allow resurrection
  • bd migrate-tombstones command

    • Converts legacy deletions.jsonl entries to inline tombstones
    • Archives old file as deletions.jsonl.migrated
  • Enhanced Git Worktree Support

    • Shared .beads database across all worktrees
    • Worktree-aware database discovery
    • Git hooks automatically adapt to worktree context
    • Documentation in docs/WORKTREES.md

Fixed

  • Multi-hyphen prefix parsing (GH#405) - Prefixes like my-cool-project now work correctly
  • bd sync on sync branch (GH#519) - Fixed sync when already on the sync branch
  • Pre-commit hook with removed .beads (GH#483) - No longer blocks commits after beads removal
  • Priority format error messages (GH#517) - Clearer guidance on P0-P4 format
  • Status naming consistency (GH#444) - Uses in_progress everywhere
  • Orphan detection with dots in directory names (GH#508) - No more false positives
  • Circular error in pre-push hook (GH#532) - Fixed infinite loop scenario
  • Doctor --fix auto-migrates tombstones - Automatic migration during repair
  • Daemon detects external DB replacement (#564) - Contributed by @deblasis
  • Import handles hierarchical hash IDs (#584) - Contributed by @rsnodgrass
  • Nested worktree detection (GH#509) - Correctly finds .beads/ in parent repo
  • Import skips cross-prefix content matches - Prevents incorrect renames
  • 3-way merge tolerates missing issues - More robust conflict resolution
  • Pre-commit warns instead of failing on flush error - Graceful degradation

Security

  • Go toolchain updated to 1.24.11 - Addresses CVEs in standard library

CI

  • PR check for .beads/issues.jsonl changes - Rejects accidental database commits

Dependencies

  • Bump github.com/ncruces/go-sqlite3 to 0.30.3
  • Bump github.com/spf13/cobra to 1.10.2
  • Bump golang.org/x/mod to 0.31.0
  • Bump golang.org/x/term to 0.38.0
  • Bump pydantic to 2.12.5 (beads-mcp)
  • Bump fastmcp to 2.14.1 (beads-mcp)

[0.29.0] - 2025-12-03

Added

  • --estimate / -e flag for bd create and bd update (GH #443)

    • Add time estimates to issues in minutes
    • Example: bd create "My task" --estimate 120 (2 hours)
    • Example: bd update bd-xyz --estimate 60 (1 hour)
    • Enables planning and prioritization features in vscode-beads
  • bd doctor improvements

    • SQLite integrity check - Detects database corruption
    • Configuration value validation - Validates config settings
    • Stale sync branch detection - Warns about abandoned beads-sync branches
    • --output flag - Export diagnostics to file for sharing
    • --dry-run flag - Preview fixes without applying
    • Per-fix confirmation mode - Approve each fix individually
  • --readonly flag - Read-only mode for worker sandboxes

    • Blocks all write operations
    • Useful for parallel worker processes that should only read

Fixed

  • bd sync safety improvements

    • Auto-push after merge with safety check
    • Handle diverged histories with content-based merge
    • Multiple safety check enhancements
  • Auto-resolve merge conflicts deterministically

    • All field conflicts resolved without prompts
    • Uses consistent rules for field-level merging
  • 3-char all-letter base36 hash support (GH #446)

    • Fixes prefix extraction for edge case hashes like "bd-abc"
  • bd ready message fix

    • Shows correct message when all issues are closed
  • Version notification spam fix

    • Store version in gitignored .local_version file
  • Nix flake vendorHash update

    • Fixed build after dependency bumps

Documentation

  • Added perles and vscode-beads to Third-Party Tools
  • Encourage batch close and parallel creation in bd prime output

[0.28.0] - 2025-12-01

Added

  • bd daemon --local flag (#433) - Run daemon without git operations

    • Ideal for multi-repo setups where git sync happens externally
    • Prevents daemon from triggering git commits/pushes
    • Use with worktrees or when beads sync is handled by another process
  • bd daemon --foreground flag - Run daemon in foreground mode

    • For systemd/supervisord/launchd integration
    • Logs to stdout instead of background file
    • Process stays attached to terminal
  • bd migrate-sync command - Migrate to sync.branch workflow

    • Moves beads data to a dedicated sync branch
    • Keeps main branch clean of .beads/ commits
    • Automated setup of sync.branch configuration

Fixed

  • Database Migration: close_reason column

    • Added missing database column for close_reason field
    • Fixes sync loops where close_reason was lost on import/export
    • Automatic migration on first run
  • Multi-repo Prefix Filtering (GH #437)

    • Issues now filtered by prefix when flushing from non-primary repos
    • Prevents issues from other projects appearing in exports
  • Parent-Child Dependency UX (GH #440)

    • Fixed backwards documentation in DEPENDENCIES.md
    • bd show now displays epic children under "Children" not "Blocks"
    • Clearer UI labels for dependency relationships
  • sync.branch Workflow Fixes

    • Fixed .beads/ restoration from branch after sync
    • Prevents final flush after sync.branch restore
    • bd doctor now detects when on sync branch
  • Jira API Migration

    • Updated from deprecated Jira API v2 to v3
    • Fixes authentication issues with newer Jira instances
  • Redundant Database Queries

    • Removed extra GetCloseReason() calls after column migration
    • Improves query performance for issue retrieval

Documentation

  • Added go install fallback instructions for Claude Code web (GH #439)
  • Added uninstall documentation (GH #445)

Internal

  • Refactored daemon sync functions to reduce ~200 lines of duplication
  • Consolidated local-only sync functions into shared implementation

[0.27.2] - 2025-11-30

Fixed

  • CRITICAL: Prevent Mass Database Deletion on JSONL Reset

    • git-history-backfill now includes safety guard to prevent purging entire database
    • If >50% of issues would be deleted via git history, operation is aborted with warning
    • Threshold prevents accidental deletion when JSONL is reset (git reset, branch switch, etc.)
    • If 10-50% would be deleted, operation proceeds but shows warning message
    • Fixes bug where git reset --hard origin/main would lose all issues
  • Fix Fresh Clone Initialization

    • bd init now works on fresh clones that have JSONL but no database
    • Auto-detects issue prefix from existing JSONL (no --prefix flag needed)
    • Prevents "database not found" errors on first run in a cloned repository
  • Import Warning for Deleted Issues

    • New warning message when issues are skipped due to deletions manifest
    • Helps users understand why expected issues aren't being imported
    • Warns user to check bd deleted for history of removed issues
  • Extract Issue Prefix for 3-char Hashes (#425)

    • ExtractIssuePrefix() now handles base36 hashes as short as 3 characters
    • Previously required 4+ chars, breaking hyphenated prefixes (e.g., document-intelligence-0sa)
    • Updated hash validation to accept base36 (0-9, a-z) instead of just hex
    • Requires at least one digit to distinguish hashes from English words

[0.27.0] - 2025-11-29

Added

  • Custom Status States: Define custom issue statuses via config

    • Configure project-specific statuses like testing, blocked, review
    • Status validation ensures only configured statuses are used
    • Backwards compatible: open/in_progress/closed always work
  • Contributor Fork Workflows: bd init --contributor now auto-configures sync.remote=upstream

    • Syncs issues from upstream rather than origin
    • Ideal for contributors working on forks
  • Git Worktree Support: Full support for git worktrees (#416)

    • bd hooks install now works correctly in worktrees
    • Helpful message when entering worktree repo without beads initialized
    • Hooks properly detect worktree vs main repository
  • Daemon Health Checks: Health monitoring in daemon event loop

    • Periodic health checks detect stale database state
    • Auto-recovery from detected inconsistencies
  • Fresh Clone Detection: bd doctor now detects fresh clones

    • Suggests bd init when JSONL exists but no database
    • Improved onboarding experience for new contributors
  • bd sync --squash: Batch multiple sync commits into one

    • Reduces commit noise when syncing frequently
    • Optional flag for cleaner git history
  • Error Handling Helpers: Extracted FatalError/WarnError utilities

    • Consistent error formatting across codebase
    • Better error messages for users

Fixed

  • CRITICAL: Sync Corruption Prevention: Multiple fixes prevent stale database from corrupting JSONL

    • Hash-based staleness detection: SHA256 hash comparison catches content mismatches
    • Reverse ZFC check: Detects when JSONL has more issues than DB
    • Stale daemon connection (eb4b81d): Prevents corruption from stale SQLite connections
    • Combined, these fixes eliminate the sync corruption bugs that affected v0.26.x
  • Multi-Hyphen Prefix Support (#419): Hash IDs with multi-part prefixes now handled correctly

    • Example: my-app-abc123 correctly parsed as prefix my-app
  • Out-of-Order Dependencies (#414): JSONL import handles dependencies before their targets exist

    • Fixes import failures when issues reference not-yet-imported dependencies
  • --from-main Sync Mode (#418): Now defaults to noGitHistory=true

    • Prevents spurious deletions when syncing from main branch
  • JSONL-Only Mode Detection: Auto-detects when config has no-db: true

    • Properly handles repositories using JSONL without SQLite
  • Init Safety Guard: Prevents overwriting existing JSONL data on init

    • Warns user when data already exists, requires confirmation
  • Snapshot Cleanup: Properly cleans up snapshot files after sync

    • Removes .beads/*.snapshot files that could cause conflicts
  • Daemon Registry Locking: Cross-process locking prevents registry corruption

    • Fixes race conditions when multiple processes access daemon registry
  • Doctor Merge Artifacts: Excludes merge artifacts from "multiple JSONL" warning

    • Reduces false positives during merge resolution

Changed

  • Documentation: Fixed birthday paradox threshold explanation in README
  • Documentation: Corrected bd dep add syntax and semantics

Community

  • PR #419: Multi-hyphen prefix support
  • PR #418: --from-main noGitHistory default
  • PR #416: Git worktree hooks support
  • PR #415: CI fixes
  • PR #414: Out-of-order dependency handling
  • PR #404: Error on invalid JSON during init (@joelklabo)

[0.26.2] - 2025-11-29

Fixed

  • Hash-Based Staleness Detection: Prevents stale DB from corrupting JSONL when counts match
    • Previous count-based check (0.26.1) missed cases where DB and JSONL had similar issue counts
    • New detection computes SHA256 hash of JSONL content and stores it after import
    • On export, compares current JSONL hash against stored hash to detect modifications
    • If JSONL was modified externally (e.g., by git pull), triggers re-import before export
    • Ensures database is always synchronized with JSONL before exporting changes

[0.26.1] - 2025-11-29

Fixed

  • CRITICAL: Reverse ZFC Check: Prevents stale database from corrupting JSONL
    • Root cause: bd sync exports DB to JSONL before pulling from remote
    • If local DB is stale (fewer issues than JSONL), stale data would corrupt the JSONL
    • Added reverse ZFC check: detects when JSONL has >20% more issues than DB
    • When detected, imports JSONL first to sync database before any export
    • Prevents fresh/stale clones from exporting incomplete database state

[0.26.0] - 2025-11-27

Added

  • bd doctor --check-health: Lightweight health checks for startup hooks (3fe94f2)

    • Quick, silent health checks (exit 0 on success, non-zero on issues)
    • Checks: version mismatch, sync.branch config, outdated hooks
    • New hints.doctor config option to suppress doctor hints globally
    • Git hooks now call bd doctor --check-health in post-merge/post-checkout
  • --no-git-history Flag: Prevent spurious deletions during import/sync (5506486)

    • Use when git history is unreliable (shallow clones, squash merges)
    • Prevents deletion manifest from removing issues based on stale history
  • gh2jsonl Hash ID Mode: Content-based ID generation for GitHub imports (#383 by @deangiberson)

    • --id-mode {sequential|hash} flag (default: sequential for backward compatibility)
    • --hash-length {3,4,5,6,7,8} for configurable hash length (default: 6)
    • Hash IDs are deterministic using title, description, creator, timestamp
  • CI Provenance Attestation: npm publish now includes provenance attestation (03d62d0)

  • bdui: Added to Third-Party Tools ecosystem (#384)

Fixed

  • Critical: MCP Protocol Stdin Fix (PR #400 by @cleak)

    • Subprocess stdin inheritance was breaking MCP JSON-RPC protocol
    • All subprocess calls now use stdin=subprocess.DEVNULL
    • Fixes hanging/blocking issues in Claude Desktop MCP integration
  • Git Worktree Staleness (#399)

    • Staleness check was failing after writes in git worktrees
    • Now uses RFC3339Nano precision for last_import_time metadata
  • Multi-Part Prefix Support (#398)

    • Issue ID extraction now correctly handles multi-part prefixes
    • Example: my-app-123 correctly extracts prefix my-app
  • bd sync Commit Scope

    • bd sync now only commits .beads/ files, not other staged files
    • Prevents accidental commits of unrelated staged changes
  • Auto-Import to Wrong File

    • Fixed auto-import exporting to wrong JSONL file
    • FindJSONLPath now correctly skips deletions.jsonl
  • Deletions.jsonl Handling

    • Git hooks now properly stage deletions.jsonl for cross-clone propagation
    • bd doctor no longer warns about deletions.jsonl
    • Prevent rebase failures from deletions.jsonl writes
  • Defense-in-Depth

    • Added additional check for --no-auto-import flag
  • Tilde Expansion: Global gitignore path now expands ~ correctly

  • beads-mcp Type Checking: Resolved all mypy type checking errors

  • CI Test Stability: Fixed Windows test failures

Changed

  • Stealth Mode: Removed global gitattributes setup from bd init --stealth (#391)

    • Stealth mode now purely local with no global git configuration
  • Pre-Push Hook Error Message: Improved clarity when sync fails (#390)

Refactoring

  • Extract path canonicalization and database search helpers (7d765c2)
  • Consolidate check-health DB access and expand hook checks (3458956)

Documentation

  • Sync skill CLI reference with current docs (62d1dc9)

Community

  • PR #400: MCP stdin fix (@cleak)
  • PR #398: Multi-part prefix support
  • PR #391: Stealth mode gitattributes removal
  • PR #390: Pre-push hook error message (@jonathanpberger)
  • PR #384: bdui ecosystem addition
  • PR #383: gh2jsonl hash ID support (@deangiberson)

[0.25.1] - 2025-11-25

Added

  • Zombie Resurrection Prevention: Stale clones can no longer resurrect deleted issues
    • New JSONL sanitization step (3.6) after git pull removes deleted issues before import
    • Prevents git's 3-way merge from re-adding issues that were deleted elsewhere
    • New bd doctor check 18: "Deletions Manifest" detects missing/empty manifest
    • New bd doctor --fix hydrates deletions.jsonl from git history for pre-v0.25.0 deletions
    • ID validation prevents false positives from non-issue JSON fields

Fixed

  • bd sync commit scope: Now commits entire .beads/ directory before pull
    • Previously only committed beads.jsonl, leaving metadata.json unstaged
    • Fixes "You have unstaged changes" error during git pull --rebase

[0.25.0] - 2025-11-25

Added

  • Deletion Propagation: Deletions now sync across clones via deletions manifest

    • New .beads/deletions.jsonl tracks deleted issues with timestamp, actor, reason
    • Import automatically purges issues that were deleted in other clones
    • Git history fallback for pruned deletion records (self-healing)
    • New bd deleted command to view deletion audit trail
    • Auto-compact during sync (opt-in via deletions.auto_compact: true)
    • Configurable retention period (deletions.retention_days, default 7)
    • Local unpushed work protected from accidental deletion
    • Full documentation in docs/DELETIONS.md
  • Stealth Mode: New bd init --stealth flag for invisible beads usage (#381)

    • Adds .beads/ to project's .gitignore automatically
    • Useful for personal issue tracking in shared repos without affecting collaborators
    • All bd functionality works normally, just not committed to git
  • Ephemeral Branch Sync: New bd sync --from-main flag

    • Syncs from main branch without pushing to remote
    • Ideal for feature branches that sync issues from main
    • Prevents pushing local branch changes to origin

[0.24.4] - 2025-11-25

Added

  • Transaction API: Full transactional support for atomic multi-operation workflows

    • New storage.Transaction interface with CreateIssue, UpdateIssue, CloseIssue, DeleteIssue
    • Dependency operations: AddDependency, RemoveDependency within transactions
    • Label operations: AddLabel, RemoveLabel with transactional semantics
    • Config/Metadata operations for atomic config+issue workflows
    • Uses BEGIN IMMEDIATE mode to prevent deadlocks
    • Automatic rollback on error or panic, commit on success
    • 1,147 lines of new implementation with comprehensive tests
  • Tip System Infrastructure: Smart contextual hints for users

    • Tips shown after successful commands (list, ready, create, show)
    • Condition-based filtering, priority ordering, probability rolls
    • Frequency limits prevent tip spam
    • Respects --json and --quiet flags
    • Deterministic testing via BEADS_TIP_SEED env var
  • Sorting for bd list and bd search: New --sort and --reverse flags

    • Sort by: priority, created, updated, closed, status, id, title, type, assignee
    • Smart defaults: priority ascending (P0 first), dates descending (newest first)
    • Works with all existing filters
  • Claude Integration Verification: New bd doctor checks

    • CheckBdInPath: verifies 'bd' is in PATH (needed for hooks)
    • CheckDocumentationBdPrimeReference: detects version mismatches in docs
  • ARM Linux Support: GoReleaser now builds for linux/arm64 (PR #371 by @tjg184)

    • Enables bd on ARM-based Linux systems like Raspberry Pi, AWS Graviton
  • Orphan Detection Migration: Identifies orphaned child issues

    • Detects issues with hierarchical IDs where parent no longer exists
    • Logs suggestions: delete, convert to standalone, or restore parent
    • Idempotent and safe to run multiple times

Fixed

  • Transaction Cache Invalidation: blocked_issues_cache now invalidates correctly

    • UpdateIssue status changes trigger cache invalidation
    • CloseIssue always invalidates (closed issues don't block)
    • AddDependency/RemoveDependency invalidate for blocking types
  • SQLITE_BUSY Retry Logic: Exponential backoff for concurrent writes

    • beginImmediateWithRetry() with 5 retries (10ms, 20ms, 40ms, 80ms, 160ms)
    • Eliminates spurious failures under normal concurrent usage
    • Context cancellation respected between retry attempts
  • bd import Argument Validation: Helpful error for common mistake

    • Running bd import file.jsonl (without -i) now shows clear error
    • Previously silently read from stdin, confusing users with "0 created"
  • ZFC Import Export Skip: Preserve JSONL source of truth

    • After stale DB import from JSONL, skip export to avoid overwriting
    • Fixes scenario where DB with fewer issues would overwrite JSONL
  • Daemon Reopen with Reason: --reason flag now works in daemon mode

    • Previously ignored in daemon RPC calls
  • Windows Test Failures: Skip file permission tests on Windows

    • Windows doesn't support Unix-style permissions (0600, 0755)
    • Core functionality still tested, only permission checks skipped

Changed

  • bd daemon UX: New --start flag, help text when no args
    • bd daemon now shows help instead of immediately starting
    • Use bd daemon --start to explicitly start
    • Auto-start still works (uses --start internally)
    • More discoverable for new users

Documentation

  • blocked_issues_cache Architecture: Document cache behavior and invalidation
  • Antivirus False Positives: Guide for handling security software alerts
  • import.orphan_handling: Complete documentation
  • Error Handling Patterns: Comprehensive audit and guidelines

Dependencies

  • Bump github.com/tetratelabs/wazero from 1.10.0 to 1.10.1 (#374)
  • Bump github.com/anthropics/anthropic-sdk-go from 1.18.0 to 1.18.1 (#373)
  • Bump actions/checkout from 4 to 6 (#372)

Community

  • PR #371: ARM Linux support (@tjg184)

[0.24.3] - 2025-11-24

Added

  • BD_GUIDE.md Generation: Version-stamped documentation for AI agents (bd-woro, 9e16469)

    • New --output flag for bd onboard command generates BD_GUIDE.md
    • Separates bd-specific instructions from project-specific instructions
    • Auto-generated header with version stamp warns against manual editing
    • Detects outdated BD_GUIDE.md and suggests regeneration after upgrades
    • Git-trackable diffs show exactly what changed between versions
  • Configurable Export Error Policies: Flexible error handling for export operations (bd-exug, e3e0a04)

    • Four policies: strict (fail-fast), best-effort (skip with warnings), partial (retry then skip), required-core (fail on core data only)
    • Per-project configuration via bd config set export.error_policy
    • Separate policy for auto-exports via auto_export.error_policy
    • Retry with exponential backoff for transient failures
    • Optional export manifests documenting completeness
  • Command Set Standardization: Complete flag and feature consistency overhaul (bd-au0, 273a4d1)

    • Global verbosity flags: --verbose/-v and --quiet/-q across all commands
    • Standardized --dry-run flag behavior across all commands
    • Label operations added to bd update: --add-labels and --remove-labels
    • Enhanced bd export with comprehensive filters (assignee, type, labels, priority, dates)
    • Enhanced bd search with date and priority filters
    • Improved documentation for bd clean vs bd cleanup disambiguation
  • Auto-Migration on Version Bump: Automatic database schema updates (bd-jgxi, 7796f5c)

    • Database version automatically synced when bd CLI is upgraded
    • Eliminates recurring "version mismatch" warnings in bd doctor
    • Best-effort and silent to avoid disrupting commands
  • Version Tracking Validation: Comprehensive bd doctor checks (bd-u4sb, d8f3eb0)

    • Validates metadata.json exists with valid LastBdVersion field
    • Warns if LastBdVersion is very old (>10 minor versions behind)
    • Ensures version tracking system is working correctly
  • Git Hooks Executable Validation: Hook reliability improvements (bd-fwul, 1c715bc)

    • bd doctor now validates git hooks have executable bit set
    • Prevents silent hook failures from incorrect permissions
  • Socket Cleanup Race Condition Fix: Daemon reliability improvement (bd-4owj, 18f8105)

    • Re-checks socket existence after lock check to avoid stale state
    • Handles daemon startup race conditions gracefully
  • Monitor Web UI Enhancements: UX/UI improvements (aa2df73)

    • Interactive stats cards as filters
    • Multi-select priority filtering with P0 support
    • Find-as-you-type search functionality
    • Modern card-based UI design with better mobile responsiveness
    • Dev mode flag for easier development
  • bump-version.sh Automation: PyPI integration (0547004)

    • New --upgrade-mcp flag for automatic beads-mcp package upgrades
    • Tries pip first, falls back to uv tool
    • Warns about version mismatch before PyPI publish

Fixed

  • JSONL Import Foreign Key Violations: Graceful handling of deletions during merge (bd-koab, d45cff5)

    • Import now continues when dependencies reference deleted issues
    • Reports skipped dependencies with clear warnings
    • Prevents complete import failure from FK constraint violations
    • Common scenario: merges that delete issues referenced by other issues
  • Metadata JSONL Path Auto-Detection: Fix configuration mismatches (bd-afd, d1641c7)

    • Auto-detects actual JSONL file when metadata.json is recreated
    • bd doctor --fix now includes DatabaseConfig() auto-repair
    • Prefers beads.jsonl over issues.jsonl (canonical name)
    • Prevents mismatches after git clean, merge conflicts, or rebases
  • JSONL Resurrection Bug: Critical fix for deleted issue resurrection (bd-v0y, c9a2e7a)

    • Removed mtime fast-path in hasJSONLChanged() causing false negatives
    • Git doesn't preserve mtime on checkout, causing incorrect change detection
    • Now always uses content hash for reliable comparison
    • Prevents bd sync from overwriting pulled JSONL and resurrecting deleted issues
  • ZFC (JSONL First Consistency): Fix stale DB overwriting JSONL on sync (bd-l0r, 1ba068f, 2e4171a, 949ab42)

    • bd sync now detects stale DB (>50% divergence from JSONL) and imports first
    • After ZFC import, skips export to prevent overwriting JSONL source of truth
    • Fixes bug where DB with 688 issues would overwrite JSONL with 62 issues after pull
    • JSONL is source of truth after git pull - DB syncs to match, not vice versa
    • Preserves local uncommitted changes while catching stale DB scenarios
  • Merge Conflict Semantics: Improved resolution policies (bd-pq5k, d4f9a05)

    • Merge logic now enforces: closed ALWAYS wins over open
    • Deletion ALWAYS wins over modification
    • Fixed closed_at handling: only set when status='closed'
    • Prevents issues from getting stuck in invalid states
    • Eliminates "zombie issues" that never die
  • JSONL Merge Conflict Auto-Resolution: Streamlined rebase workflow (bd-cwmt, 3cf5e26)

    • bd sync now auto-resolves JSONL conflicts during rebase
    • Detects rebase state and JSONL-only conflicts
    • Auto-exports from DB and continues rebase automatically
    • Eliminates manual conflict resolution in common scenarios
  • Staleness Check Error Handling: Better metadata validation (bd-2q6d, bd-o4qy, bd-n4td, b75914b)

    • CheckStaleness now returns errors for corrupted last_import_time metadata
    • Enhanced warning messages when staleness check fails
    • Handles empty string metadata (memory store behavior)
    • Prevents silent failures with corrupted metadata
  • N+1 Query Pattern in Export: Dramatic performance improvement (bd-rcmg, 9c6b375)

    • Added batch methods: GetCommentsForIssues() and GetLabels() in bulk
    • Reduced query count from ~201 to ~3-5 for 100 issues
    • Eliminated per-issue loops in handleExport() and triggerExport()
  • Sync Branch Auto-Configuration: Prevent bd sync failures after init (bd-flil, bd-rsua, a4c38d5, 83609d5)

    • bd init now auto-sets sync.branch to current git branch
    • bd doctor detects missing sync.branch config and provides --fix
    • All branch detection uses 'git symbolic-ref' to work in fresh repos
    • Fixes 'bd sync --status' error after fresh bd init
  • Merge Driver Auto-Repair: Fix stale git configurations (bd-3sz0, bd-tbz3, 1c8dd49)

    • Detects old bd versions (<0.24.0) with invalid %L/%R placeholders
    • Auto-repairs during bd init and bd doctor --fix
    • Git only supports %O (base), %A (current), %B (other) placeholders
    • Supports both canonical (issues.jsonl) and legacy (beads.jsonl) filenames
  • Unvalidated Dependency Parsing: Prevent empty ID lookups (bd-ia8r, e8a752e)

    • Validates dependsOnID is non-empty before setting discoveredFromParentID
    • Ensures parent issue exists before generating child IDs in direct mode
  • Windows CI Test Failures: Cross-platform reliability (153f724)

    • Fixed file locking issue in TestNewSQLiteStorage (added defer store.Close())
    • Fixed nil slice return in TestFindAllDatabases (explicit empty slice initialization)
    • Both issues related to stricter Windows file locking behavior
  • Invalid closed_at States: Data integrity fix (bd-1rh, b428254)

    • Removed invalid closed_at timestamps for open issues
    • Prevents inconsistent state from merge conflicts
  • Security: File permission hardening (b6870de, ae5a4ac)

    • Changed file permissions from 0644 to 0600 for security (gosec G302)
    • Config files now owner read/write only (not world readable)
    • Added comprehensive security tests for WriteFile permissions
    • Handles read-only files by fixing permissions before writing

Changed

  • bd init Defaults: Better out-of-box experience (bd-bxha, ec4117d)

    • Git hooks and merge driver now installed by default
    • Removed interactive prompts (simpler workflow)
    • New opt-out flags: --skip-hooks and --skip-merge-driver
    • Shows warning messages on failure with suggestion to run bd doctor --fix
  • bd init Validation: Automatic setup verification (bd-zwtq, 3a36d0b)

    • Runs bd doctor diagnostics at end of bd init
    • Immediately identifies configuration problems before user encounters them
    • Catches: missing hooks, unconfigured merge driver, missing docs, metadata issues
  • Internal Code Organization: Improved maintainability (bd-0a43, 58f37d0)

    • Split monolithic sqlite.go (1050 lines) into focused files
    • store.go: Database initialization and utilities
    • queries.go: Issue CRUD operations
    • config.go: Configuration and metadata
    • comments.go: Comment operations
    • Zero functional changes, all tests pass
  • Documentation: Organization improvements (a930fa3, 62b5f53)

    • Moved event-driven daemon details to docs/DAEMON.md
    • Added comprehensive error handling guidelines
    • Reduced duplication in AGENTS.md

Testing

  • Blocked Issues Cache Validation: Comprehensive cache tests (bd-13gm, 0e6ed91)

    • 8 tests verify cache invalidation behavior
    • Tests for dependency add/remove, status changes, transitive blocking
    • Direct cache table queries validate implementation correctness
  • Sync Test Optimization: Reduced boilerplate (bd-ktng, dfcbb7d)

    • Added shared git repo setup helpers
    • Refactored 19 test functions
    • Reduced duplicate code by ~300 lines

Performance

  • Daemon Log Rotation: Production-ready configuration (bd-t7ds, f454b3d)
    • Max size increased: 10MB → 50MB per file
    • Max backups increased: 3 → 7 files
    • Max age increased: 7 → 30 days
    • Better handles long-running daemons with high log output

[0.24.2] - 2025-11-22

Fixed

  • Test Stability: Complete rootCtx initialization fix for all hanging tests (issue #355, b8db5ab)

    • Fixed TestGetAssignedStatus missing rootCtx initialization (a517ec9)
    • Prevents test hangs from uninitialized context
    • Improved test reliability and isolation
  • JSONL Configuration: Improved bd doctor JSONL checks to focus on real problems (87ee3a6)

    • Reduces false positives in JSONL validation
    • Better detection of actual configuration issues

Changed

  • JSONL Filename Default: Changed default JSONL filename from beads.jsonl to issues.jsonl (c4c5c80)
    • Updated TestFindJSONLPathDefault to match new default (5eefec7)
    • Removed stale issues.jsonl in favor of configured beads.jsonl (d918e47)
    • More intuitive default filename for new users

[0.24.1] - 2025-11-22

Added

  • bd search: Date and priority filters (787fb4e)

    • --created-after, --created-before for date filtering
    • --priority-min, --priority-max for priority range filtering
    • Enables more precise search queries
  • bd count: New command for counting and grouping issues (d7f4189)

    • Count issues by status, priority, type, or labels
    • Helpful for generating statistics and reports
  • Test Infrastructure: Automatic skip list for tests (0040e80)

    • Improves test reliability and maintenance
    • Automatically manages flaky or environment-specific tests

Fixed

  • Test Stability: Fixed hanging tests by initializing rootCtx (822baa0, bd-n25)

    • Prevents test hangs from context cancellation issues
    • Better test isolation and cleanup
  • Git Merge Driver: Corrected placeholders from %L/%R to %A/%B (ddd209e)

    • Fixes merge driver configuration for proper conflict resolution
    • Uses correct git merge driver variable names
  • Database Paths: Deduplicate database paths when symlinks present (#354, f724b61)

    • Prevents duplicate database detection when symlinks are involved
    • Improves reliability in complex filesystem setups

Changed

  • bd list: Accept both integer and P-format for priority flags (2e2b8d7)

    • --priority 1 and --priority P1 now both work
    • More flexible CLI input for priority filtering
  • bd update: Added --body flag as alias for --description (bb5a480)

    • More intuitive flag name for updating issue descriptions
    • Both flags work identically for backward compatibility
  • bd update: Added label operations (3065db2)

    • --add-labels and --remove-labels flags
    • Simplifies label management in update operations
  • GitHub Copilot Support: Added .github/copilot-instructions.md (605fff1)

    • Provides project-specific guidance for GitHub Copilot
    • Improves AI-assisted development experience
  • Documentation: Moved design/audit docs from cmd/bd to docs/ (ce433bb)

    • Better organization of project documentation
    • Clearer separation of code and documentation

Performance

  • Test Suite: Deleted 7 redundant tests from main_test.go (fa727c7)

    • 3x speedup in test execution
    • Improved CI/CD performance
  • Test Coverage: Tagged 16 slow integration tests with build tags (8290243)

    • Faster local test runs with -short flag
    • CI can still run full test suite

Testing

  • Security Tests: Added security and error handling tests for lint warnings (74f3844)

    • Improved code quality and safety
    • Better coverage of edge cases
  • Shared Database Pattern: Refactored multiple test files to use shared DB pattern

    • compact_test.go, integrity_test.go, validate_test.go, epic_test.go, duplicates_test.go
    • Improved test consistency and maintainability
    • Faster test execution through better resource sharing

[0.24.0] - 2025-11-20

Added

  • bd doctor --fix: Automatic repair functionality (bd-ykd9, 7806937)

    • Automatically fixes issues detected by bd doctor
    • Repairs common database inconsistencies without manual intervention
  • bd clean: Remove temporary merge artifacts (e8355c2)

    • Cleans up .base, .ours, .theirs snapshot files
    • Helps maintain clean .beads/ directory after merges
  • bd cleanup: Enhanced bulk deletion command

    • Delete multiple closed issues efficiently
    • Improved from previous versions with better performance
  • .beads/README.md Generation: Auto-generated during bd init (bd-m7ge, e1c8853)

    • Provides project-specific beads documentation
    • Helps new contributors understand the setup
  • blocked_issues_cache Table: Performance optimization for GetReadyWork (62c1f42, ed23f8f)

    • Caches blocked issue relationships
    • Dramatically improves bd ready performance on large databases
  • Commit Hash in Version Output: Enhanced version reporting (bd-hpt5, 7c96142)

    • bd version now shows git commit hash
    • Helps identify exact build for debugging
  • Auto-detection of Issue Prefix: Scans git history to detect prefix (#277, 8f37904)

    • Automatically discovers project's issue prefix
    • Reduces manual configuration needed
  • external_ref Support in Daemon RPC: Full daemon mode support (#304, 57b6ea6)

    • MCP server can now set external references in daemon mode
    • Parity with CLI functionality
  • Context Optimization Features: AI agent improvements (#297, f7e80dd)

    • Context propagation with graceful cancellation (bd-rtp, bd-yb8, bd-2o2, 57253f9)
    • Better memory management for long-running agent sessions

Fixed

  • Critical: Auto-import Resurrection Bug (bd-khnb, 0020eb4, e28e3ea, 7b6370f)

    • Fixed critical bug where deleted issues were resurrected during auto-import
    • Cleaned up 497+ resurrected issues from production database
    • Prevents data corruption from improper JSONL replay
  • Critical: bd sync Auto-resolves Conflicts (bd-ca0b, a1e5075)

    • bd sync now automatically resolves conflicts instead of failing
    • Dramatically improves multi-agent workflow reliability
    • Eliminates manual conflict resolution in most cases
  • Critical: Content-based Timestamp Skew Prevention (bd-lm2q, d0e7047)

    • Fixed false-positive "JSONL is newer than database" warnings
    • Uses content-based comparison instead of timestamp-only
    • Prevents unnecessary imports that would corrupt state
  • Critical: bd sync DB Changes After Import (81c741b)

    • Ensures database changes are properly applied after import
    • Fixes desync issues between JSONL and database
  • Critical: Context Propagation Lifecycle Bugs (bd-rtp, bd-yb8, bd-2o2, 57253f9, a17e4af)

    • Fixed multiple context propagation issues causing crashes
    • Graceful cancellation support for long-running operations
    • Improved stability for AI agent workflows
  • Critical: Race Condition in Auto-flush (bd-52, a9b2f9f)

    • Fixed race condition in auto-flush mechanism
    • Prevents data loss during concurrent operations
  • Critical: Resource Leaks and Error Handling (#327, fb65163)

    • Fixed critical resource leaks in daemon mode
    • Improved error handling throughout codebase
  • Critical: In-memory Database Deadlock (bd-yvlc, 944ed10)

    • Fixed deadlock in migrations when using in-memory database
    • Improves test reliability
  • MCP Schema Generation Recursion Bug (GH#346, f3a678f)

    • Fixed infinite recursion in MCP schema generation
    • Prevents stack overflow crashes
  • FK Constraint Failures (bd-5arw, 345766b)

    • Fixed foreign key constraint failures in AddComment and ApplyCompaction
    • Improved data integrity
  • --parent Flag Behavior (b9919fe)

    • Now correctly creates parent-child dependency relationships
    • Previously was creating wrong dependency type
  • Exact ID Matching Priority (gh-316, 934ae04)

    • Prefers exact ID matches over prefix matches
    • Prevents ambiguous ID resolution
  • Daemon Lifetime on macOS (GH#278, 68f9bef)

    • Fixed daemon exiting after 5s on macOS due to PID 1 parent monitoring
    • Daemon now runs reliably on macOS
  • Daemon Export/JSONL Sync (GH#301, #321, 04a1996)

    • Fixed daemon export leaving JSONL newer than database
    • Ensures proper sync between export and database state
  • bd doctor Hash ID Detection (GH#322, 8c1f865)

    • Fixed doctor incorrectly diagnosing hash IDs as sequential
    • Improved detection logic for ID format validation
  • ResolvePartialID Handling (GH#336, 4432af0)

    • Improved ResolvePartialID / ResolveID handling for bd show
    • Better partial ID matching and error messages
  • bd sync Windows Upstream Detection (#281, 1deaad1)

    • Fixed upstream branch detection on Windows
    • Improved cross-platform compatibility
  • Compact Command Daemon Mode (#294, d9904a8)

    • Fixed compact command failing with 'SQLite DB needed' error when daemon running
    • Removed premature store check, uses ensureDirectMode
  • DB mtime Update After Import (#296, 9dff345)

    • Fixed DB mtime not being updated after import with 0 changes
    • Prevents false staleness warnings
  • FOREIGN KEY Constraint on Non-existent Issues (09666b4)

    • Fixed constraint failures when operating on non-existent issues
    • Better error handling and validation
  • Monitor WebUI Daemon Detection (e36baee)

    • Fixed monitor-webui failure to detect running daemon
    • Improved daemon health checking
  • Onboard Test Deadlock on Windows (4e22214)

    • Fixed deadlock in onboard tests on Windows
    • Improved test stability
  • Windows Concurrent Issue Creation (4cd26c8)

    • Fixed concurrent issue creation failures on Windows
    • Better file locking on Windows
  • Missing Git Hook Message (#306, 92f3af5)

    • Improved messaging when git hooks are missing
    • Clearer instructions for users
  • Prefix Detection for Hyphenated Apps (83472ac, bd-fasa)

    • Fixed prefix detection to only use first hyphen
    • Handles hyphenated application names correctly
  • External Ref Migration Failures (8be792a)

    • Fixed external_ref migration failure on old databases
    • Backward compatibility improvements
  • Duplicate Function Declaration (#328, 167ab67)

    • Fixed compilation failure from duplicate computeJSONLHash declaration
    • Removed old version, kept simpler implementation
    • Updated test to match new API

Changed

  • Performance Improvements (#319, 690c73f):

    • Optimized GetReadyWork to use blocked_issues_cache (ed23f8f)
    • Replaced N+1 label queries with bulk fetch in bd list (968d9e2)
    • Cache invalidation for blocked_issues_cache (614ba8a)
    • Significant speedup for large databases
  • FlushManager Improvements (445857f)

    • Added constants for magic numbers
    • Enhanced error logging
    • Comprehensive functional tests
  • Auto-upgrade .beads/.gitignore (#300, f4a2f87)

    • Automatically upgrades .gitignore on bd operations
    • Ensures latest patterns are always applied
  • Code Refactoring:

    • Extract duplicated validation logic to internal/validation (d5239ee)
    • Centralize error handling patterns in storage layer (bd-bwk2, 3b2cac4)
    • Extract duplicated validation and flag logic (bd-g5p7, bbfedb0)
    • Improved code organization and maintainability
  • Documentation Improvements:

    • Document files created by bd init and clarify .gitattributes (721274b, e7fd1dd)
    • How to resolve merge conflicts in .beads/beads.jsonl (4985a68)
    • Document MCP tools loading issue in Claude Code (GH#346, 79b8dbe)
    • Add uv prerequisite to Claude Code plugin docs (#293, a020c6c)
    • Don't auto-install Go in Windows installer (#302, 0cba73b)
  • Improved Error Messages (#349, 27c0c33)

    • Compact error messages
    • Remove bogus merge suggestion
    • Add daemon/maintenance docs
  • AGENTS.md Refactoring (21a0656)

    • Extracted detailed instructions to prevent context pollution
    • Better organization for AI agent consumption
  • Type Safety Improvements (9e57cb6)

    • Improved type safety in beads-mcp
    • Fixed minor type issues
  • Test Improvements:

    • Fix CI regressions and stabilize tests (7b63b5a)
    • Fix parallel test deadlock (1fc9bf6)
    • Annotate gosec-safe file accesses (bf9b2c8)
  • Local-only Git Repo Support (bd-biwp, 4de9f01)

    • Support repositories without remote origin
    • Better handling of local development workflows
  • Version Marker in Post-checkout Hook (ad2154b)

    • Add version marker to post-checkout hook
    • Include in CheckGitHooks for better version tracking

Performance

  • GetReadyWork Optimization (bd-5qim, 690c73f, 62c1f42, ed23f8f)

    • Introduced blocked_issues_cache table
    • Eliminated expensive recursive queries
    • Dramatically faster for large dependency graphs
  • bd list N+1 Query Elimination (968d9e2)

    • Replaced per-issue label queries with bulk fetch
    • Significant speedup when listing many labeled issues

Community

  • Pull Requests:

    • #338: Prevent daemon from exiting when launcher process exits (@cpdata)
    • #337: Improve ResolvePartialID handling (@cpdata)
    • #333: Fix doctor incorrectly diagnosing hash IDs (@cpdata)
    • #327: Address critical resource leaks and error handling
    • #306: Improve missing git hook message
    • #304: Add external_ref support to daemon mode RPC
    • #302: Windows installer improvements
    • #300: Automatic .beads/.gitignore upgrade
    • #297: Context optimization features for AI agents
    • #296: Fix DB mtime update after import
    • #294: Fix compact command in daemon mode
    • #293: Add uv prerequisite documentation
    • #281: Fix bd sync Windows upstream detection
    • #277: Auto-detection of issue prefix from git history
  • Dependency Updates:

    • Bump github.com/anthropics/anthropic-sdk-go from 1.17.0 to 1.18.0 (#330)
    • Bump golang.org/x/mod from 0.29.0 to 0.30.0 (#331)
    • Bump fastmcp from 2.13.0.2 to 2.13.1 (#332)
    • Bump pydantic from 2.12.0 to 2.12.4 (#285)
    • Bump pydantic-settings from 2.11.0 to 2.12.0 (#286)
    • Bump golangci/golangci-lint-action from 8 to 9 (#287)
    • Bump golang.org/x/sys from 0.36.0 to 0.38.0 (#288)
    • Bump github.com/ncruces/go-sqlite3 from 0.29.1 to 0.30.1 (#290)
    • Bump github.com/google/go-cmp from 0.6.0 to 0.7.0 (#291)

Notes

This release represents a major stability and performance improvement with 179 commits since 0.23.1. Key themes:

  • Reliability: Fixed critical auto-import resurrection bug and multiple daemon issues
  • Performance: Significant optimizations for bd ready and bd list
  • AI Agent Support: Improved context propagation and error handling
  • Cross-platform: Better Windows and macOS support
  • Developer Experience: Auto-detection, better error messages, improved docs

[0.23.1] - 2025-11-08

Fixed

  • #263: Database mtime not updated after import causing false bd doctor warnings

    • When bd sync --import-only completed, SQLite WAL mode wouldn't update the main database file's mtime
    • This caused bd doctor to incorrectly warn "JSONL is newer than database" even when perfectly synced
    • Now updates database mtime after imports to prevent false warnings
  • #261: SQLite URI missing 'file:' prefix causing version detection failures

    • Without 'file:' scheme, SQLite treated ?mode=ro as part of filename instead of connection option
    • Created bogus files like beads.db?mode=ro
    • Caused bd doctor to incorrectly report "version pre-0.17.5 (very old)" on modern databases
  • bd-17d5: Conflict marker false positives on JSON-encoded content

    • Issues containing JSON strings with <<<<<<< would trigger false conflict marker detection
    • Now checks raw bytes before JSON decoding to avoid false positives
  • bd-ckvw: Schema compatibility probe prevents silent migration failures

    • Migrations could fail silently, causing cryptic "no such column" and UNIQUE constraint errors later
    • Now probes schema after migrations, retries once if incomplete, and fails fast with clear error
    • Daemon refuses RPC if client has newer minor version to prevent schema mismatches
  • #264/#262: Remove stale --resolve-collisions references

    • Docs/error messages still referenced --resolve-collisions flag (removed in v0.20)
    • Fixed post-merge hook error messages and git-hooks README

Changed

  • bd-auf1: Auto-cleanup snapshot files after successful merge

    • .beads/ no longer accumulates orphaned .base, .ours, .theirs snapshot files after merges
  • bd-ky74: Optimize CLI tests with in-process testing

    • Converted exec.Command() tests to in-process rootCmd.Execute() calls
    • Dramatically faster: 10+ minutes → just a few seconds
    • Improved test coverage from 20.2% to 23.3%
  • bd-6uix: Message system improvements

    • 30s HTTP timeout prevents hangs, full message reading, --importance validation, server-side filtering
  • Remove noisy version field from metadata.json

    • Eliminated redundant version mismatch warnings on every bd upgrade
    • Daemon version checking via RPC is sufficient

Added

  • Go agent example with Agent Mail support
  • Agent Mail multi-workspace deployment guide and scripts

[0.23.0] - 2025-11-08

Added

  • Agent Mail Integration: Complete Python adapter library with comprehensive documentation and multi-agent coordination tests

    • Python adapter library in integrations/agent-mail-python/
    • Agent Mail quickstart guide and comprehensive integration docs
    • Multi-agent race condition tests and failure scenario tests
    • Automated git traffic benchmark showing 98.5% reduction in git traffic compared to git-only sync
    • Bash-agent integration example
  • bd info --whats-new: Agent version awareness for quick upgrade summaries

    • Shows last 3 versions with workflow-impacting changes
    • Supports --json flag for machine-readable output
    • Helps agents understand what changed without re-reading full docs
  • bd hooks install: Embedded git hooks command

    • Replaces external install script with native command
    • Git hooks now embedded in bd binary
    • Works for all bd users, not just source repo users
  • bd cleanup: Bulk deletion command for closed issues

    • Agent-driven compaction for large databases
    • Removes closed issues older than specified threshold

Fixed

  • 3-way JSONL Merge: Auto-invoked on conflicts

    • Automatically triggers intelligent merge on JSONL conflicts
    • No manual intervention required
    • Warning message added to zombie issues.jsonl file
  • Auto-import on Missing Database (ab4ec90): bd import now auto-initializes database when missing

  • Daemon Crash Recovery: Panic handler with socket cleanup prevents orphaned processes

  • Stale Database Exports: ID-based staleness detection prevents exporting stale data

  • Windows MCP Subprocess Timeout: Fix for git detection on Windows

  • Daemon Orphaning (a6c9579): Track parent PID and exit when parent dies

  • Test Pollution Prevention (bd-z528, bd-2c5a): Safeguards to prevent test issues in production database

  • Client Self-Heal (a236558): Auto-recovery for stale daemon.pid files

  • Post-Merge Hook Error Messages (abb1d1c): Show actual error messages instead of silent failures

  • Auto-import During Delete: Disable auto-import during delete operations to prevent conflicts

  • MCP Workspace Context: Auto-detect workspace from CWD

  • Import Sync Warning: Warn when import syncs with working tree but not git HEAD

  • GH#254: bd init now detects and chains with existing git hooks

  • GH#249: Add nil storage checks to prevent RPC daemon crashes

  • GH#252: Fix SQLite driver name mismatch causing "unknown driver" errors

  • Nested .beads Directories: Prevent creating nested .beads directories

  • Windows SQLite Support: Fix SQLite in releases for Windows

Changed

  • Agent Affordances (observations from agents using beads):

    • bd new: Added as alias for bd create command (agents often tried this)
    • bd list: Changed default to one-line-per-issue format to prevent agent miscounting; added --long flag for previous detailed format
  • Developer Experience:

    • Extracted supplemental docs from AGENTS.md for better organization
    • Added warning for working tree vs git HEAD sync mismatches
    • Completion commands now work without database
    • Config included in bd info JSON output
    • Python cache files added to .gitignore
    • RPC diagnostics available via BD_RPC_DEBUG env var
    • Reduced RPC dial timeout from 2s to 200ms for fast-fail
    • Standardized daemon detection with tryDaemonLock probe
    • Improved internal/daemon test coverage to 60%
  • Code Organization:

    • Refactored snapshot management into dedicated module
    • Documented external_ref in content hash behavior
    • Added MCP server functions for repair commands
    • Added version number to beads-mcp startup log
    • Added system requirements section for glibc compatibility in docs
  • Release Automation:

    • Automatic Homebrew formula update in release workflow
    • Gitignore Formula/bd.rb (auto-generated, real source is homebrew-beads tap)
  • Other:

    • Added docs/ directory to links (#242)
    • RPC monitoring solution with web UI as implementation example (#244)
    • Remove old install.sh script, replaced by bd hooks install
    • Remove vc.db exclusion from FindDatabasePath filter

[0.22.1] - 2025-11-06

Added

  • Vendored beads-merge by @neongreen: Native bd merge command for intelligent JSONL merging

  • Git Hook Version Detection (bd-iou5, 991c624): bd info now detects outdated git hooks

    • Adds version markers to all git hook templates (pre-commit, post-merge, pre-push)
    • Warns when installed hooks are outdated or missing
    • Suggests running examples/git-hooks/install.sh to update
    • Prevents issues like the --resolve-collisions flag error after updates
  • Public API for External Extensions (8f676a4): Extensibility improvements for third-party tools

  • Multi-Repo Patterns Documentation (e73f89e): Comprehensive guide for AI agents working across multiple repositories

  • Snapshot Versioning (a891ebe): Add versioning and timestamp validation for snapshots

  • --clear-duplicate-external-refs flag for bd import command (9de98cf)

Fixed

  • Multi-Workspace Deletion Tracking (708a81c, e5a6c05, 4718583): Proper deletion tracking across multiple workspaces
    • Fixes issue where deletions in one workspace weren't propagated to others
    • Added DeleteIssue to Storage interface for backend extensibility (e291ee0)
  • Import/Export Deadlock (a0d24f3): Prevent import/export from hanging when daemon is running
  • Pre-Push Hook (3ba245e): Fix pre-push hook blocking instead of exporting
  • Hash ID Recognition (c924731, 055f1d9): Fix isHashID to recognize Base36 hash IDs and IDs without a-f letters
  • Git Merge Artifacts (41b1a21): Ignore merge artifacts in .beads/.gitignore
  • bd status Command (1edf3c6): Now uses git history for recent activity detection
  • Performance: Add raw string equality short-circuit before jsonEquals (5c1f441)

Changed

  • Code Organization:

    • Extract SQLite migrations into separate files (b655b29)
    • Centralize BD_DEBUG logging into internal/debug package (95cbcf4)
    • Extract normalizeLabels to internal/util/strings.go (9520e7a)
    • Reorganize project structure: move Go files to internal/beads, docs to docs/ (584c266)
    • Remove unused internal/daemonrunner/ package (~1,500 LOC) (a7ec8a2)
  • Testing:

    • Optimize test suite with testing.Short() guards for faster local testing (11fa142, 0f4b03e)
    • Add comprehensive tests for merge driver auto-config (6424ebd)
    • Add comprehensive tests for 3-way merge functionality (14b2d34)
    • Add edge case tests for getMultiRepoJSONLPaths() (78c9d74)
  • CI/CD:

    • Separate Homebrew update workflow with PAT support (739786e)
    • Add manual trigger to Homebrew workflow for testing (563c12b)
    • Fix Linux checksums extraction in Homebrew workflow (c47f40b)
    • Add script to automate Nix vendorHash updates (#235)

Performance

  • Cache getMultiRepoJSONLPaths() to avoid redundant calls (7afb143)

[0.22.0] - 2025-11-05

Added

  • Intelligent Merge Driver (bd-omx1, 52c5059): Auto-configured git merge driver for JSONL conflict resolution

    • Vendors beads-merge algorithm for field-level 3-way merging
    • Automatically configured during bd init (both interactive and --quiet modes)
    • Matches issues by identity (id + created_at + created_by)
    • Smart field merging: timestamps→max, dependencies→union, status/priority→3-way
    • Eliminates most git merge conflicts in .beads/beads.jsonl
  • Onboarding Wizards (b230a22): New bd init workflows for different collaboration models

    • bd init --contributor: OSS contributor wizard (separate planning repo)
    • bd init --team: Team collaboration wizard (branch-based workflow)
    • Interactive setup with fork detection and remote configuration
    • Auto-configures sync settings for each workflow
  • Migration Tools (349817a): New bd migrate-issues command for cross-repo issue migration

    • Migrate issues between repositories while preserving dependencies
    • Source filtering (by label, priority, status, type)
    • Automatic remote repo detection and push
    • Complete multi-repo workflow documentation
  • Multi-Phase Development Guide (3ecc16e): Comprehensive workflow examples

    • Multi-phase development (feature → integration → deployment)
    • Multiple personas (designer, frontend dev, backend dev)
    • Best practices for complex projects
  • Dependency Status (3acaf1d): Show blocker status in bd show output

    • Displays "Blocked by N open issues" when dependencies exist
    • Shows "Ready to work (no blockers)" when unblocked
  • DevContainer Support (247e659): Automatic bd setup in GitHub Codespaces

    • Pre-configured Go environment with bd pre-installed
    • Auto-detects existing .beads/ and imports on startup
  • Session Completion Protocol (095e40d): Session-ending checklist for AI agents

    • Quality gates, sync procedures, git cleanup
    • Ensures clean handoff between sessions

Fixed

  • SearchIssues N+1 Query (bd-5ots, e90e485): Eliminated N+1 query bug in label loading

    • Batch-loads labels for all issues in one query
    • Significant performance improvement for bd list with many labeled issues
  • Sync Validation (bd-9bsx, 5438485): Prevent infinite dirty loop in auto-sync

    • Added export verification to detect write failures
    • Ensures JSONL line count matches database after export
  • bd edit Direct Mode (GH #227, d4c73c3): Force bd edit to always use direct mode

    • Prevents daemon interference with interactive editor sessions
    • Resolves hang issues when editing in terminals
  • SQLite Driver on arm64 macOS (f9771cd): Fixed missing SQLite driver in arm64 builds

    • Explicitly imports CGO-enabled sqlite driver
    • Resolves "database driver not found" errors on Apple Silicon
  • external_ref Type Handling (e1e58ef): Handle both string and *string in UpdateIssue RPC

    • Fixes type mismatch errors in MCP server
    • Ensures consistent API behavior
  • Windows Test Stability (2ac28b0, 8c5e51e): Skip flaky concurrent tests on Windows

    • Prevents false failures in CI/CD
    • Improves overall test suite reliability

Changed

  • Test Suite Performance (0fc4da7): Optimized test suite for 15-18x speedup

    • Reduced redundant database operations
    • Parallelized independent test cases
    • Faster CI/CD builds
  • Priority Format (b8785d3): Added support for P-prefix priority format (P0-P4)

    • Accepts both --priority 1 and --priority P1
    • More intuitive for GitHub/Jira users
  • --label Alias (85ca8c3): Added --label as alias for --labels in bd create

    • Both singular and plural forms now work
    • Improved CLI ergonomics
  • --parent Flag in Daemon Mode (fc89f15): Added --parent support in daemon RPC

    • MCP server can now set parent relationships
    • Parity with CLI functionality

Documentation

  • Multi-Repo Migration Guide (9e60ed1): Complete documentation for multi-repo workflows

    • OSS contributors, teams, multi-phase development
    • Addresses common questions about fork vs branch workflows
  • beads-merge Setup Instructions (527e491): Enhanced merge driver documentation

    • Installation guide for standalone binary
    • Jujutsu configuration examples

[0.21.9] - 2025-11-05

Added

  • Epic/Child Filtering (bd-zkl, fbe790a): New bd list filters for hierarchical issue queries

    • --ancestor <id>: Filter by ancestor issue (shows all descendants)
    • --parent <id>: Filter by direct parent issue
    • --epic <id>: Alias for --ancestor (more intuitive for epic-based workflows)
    • ancestor_id field added to issue type for efficient epic hierarchy queries
  • Advanced List Filters: Pattern matching, date ranges, and empty checks

    • Pattern matching: --title-contains, --desc-contains, --notes-contains (case-insensitive substring)
    • Date ranges: --created-after/before, --updated-after/before, --closed-after/before
    • Empty checks: --empty-description, --no-assignee, --no-labels
    • Priority ranges: --priority-min, --priority-max
  • Database Migration (bd-bb08, 3bde4b0): Added ON DELETE CASCADE to child_counters table

    • Prevents orphaned child counter records when issues are deleted
    • Comprehensive migration tests ensure data integrity

Fixed

  • Import Timestamp Preservation (8b9a486): Fixed critical bug where closed_at timestamps were lost during sync

    • Ensures closed issues retain their original completion timestamps
    • Prevents issue resurrection timestamps from overwriting real closure times
  • Import Config Respect (7292c85): Import now respects import.missing_parents config setting

    • Previously ignored config for parent resurrection behavior
    • Now correctly honors user's preference for handling missing parents
  • GoReleaser Homebrew Tap (37ed10c): Fixed homebrew tap to point to steveyegge/homebrew-beads

    • Automated homebrew formula updates now work correctly
    • Resolves brew installation issues
  • npm Package Versioning (626d51d): Added npm-package to version bump script

    • Ensures @beads/bd npm package stays in sync with CLI releases
    • Prevents version mismatches across distribution channels
  • Linting (52cf2af): Fixed golangci-lint errors

    • Added proper error handling
    • Added gosec suppressions for known-safe operations

Changed

  • RPC Filter Parity (510ca17): Comprehensive test coverage for CLI vs RPC filter behavior
    • Ensures MCP server and CLI have identical filtering semantics
    • Validates all new filters work correctly in both modes

[0.21.8] - 2025-11-05

Added

  • Parent Resurrection: Automatic resurrection of deleted parent issues from JSONL history
    • Prevents import failures when parent issues have been deleted
    • Creates tombstone placeholders for missing hierarchical parents
    • Best-effort dependency resurrection from JSONL

Changed

  • Error Messages: Improved error messages for missing parent issues
    • Old: "parent issue X does not exist"
    • New: "parent issue X does not exist and could not be resurrected from JSONL history"
    • Breaking: Scripts parsing exact error messages may need updates

Fixed

  • JSONL Resurrection Logic: Fixed to use LAST occurrence instead of FIRST (append-only semantics)
  • Version Bump Script: Added --tag and --push flags to automate release tagging
    • Addresses confusion where version bump doesn't trigger GitHub release
    • New usage: ./scripts/bump-version.sh X.Y.Z --commit --tag --push

[0.21.7] - 2025-11-04

Fixed

  • Memory Database Connection Pool: Fixed :memory: database handling to use single shared connection

    • Prevents "no such table" errors when using in-memory databases
    • Ensures connection pool reuses the same in-memory instance
    • Critical fix for event-driven daemon mode tests
  • Test Suite Stability: Fixed event-driven test flakiness

    • Added waitFor helper for event-driven testing
    • Improved timing-dependent test reliability

[0.21.6] - 2025-11-04

Added

  • npm Package: Created @beads/bd npm package for Node.js/Claude Code for Web integration

    • Native binary downloads from GitHub releases
    • Integration tests and release documentation
    • Postinstall script for platform-specific binary installation
  • Template Support: Issue creation from markdown templates

    • Create multiple issues from a single file
    • Structured format for bulk issue creation
  • bd comment Alias: Convenient shorthand for bd comments add

Changed

  • Base36 Issue IDs (GH #213): Switched from hex to Base36 encoding for shorter, more readable IDs
    • Reduces ID length while maintaining uniqueness
    • More human-friendly format

Fixed

  • SQLite URI Handling: Fixed file:// URI scheme to prevent query params in filename

    • Prevents database corruption from malformed URIs
    • Fixed :memory: database connection strings
  • bd init --no-db Behavior (GH #210): Now correctly creates metadata.json and config.yaml

    • Previously failed to set no-db: true flag
    • Improved metadata-only initialization workflow
  • Symlink Path Resolution: Fixed findDatabaseInTree to properly resolve symlinks

  • Epic Hierarchy Display: Fixed bd show command to correctly display epic child relationships

  • CI Stability: Fixed performance thresholds, test eligibility, and lint errors

Dependencies

  • Bumped github.com/anthropics/anthropic-sdk-go from 1.14.0 to 1.16.0
  • Bumped fastmcp from 2.13.0.1 to 2.13.0.2

[0.21.5] - 2025-11-02

Fixed

  • Critical Double JSON Encoding Bug (bd-1048, bd-4ec8): Fixed widespread bug in daemon RPC calls where ResolveID responses were incorrectly converted using string(resp.Data) instead of json.Unmarshal. This caused IDs to become double-quoted ("\"bd-1048\"") and database lookups to fail. Affected commands:

    • bd show - nil pointer dereference and 3 instances of double encoding
    • bd dep add/remove/tree - 5 instances
    • bd label add/remove/list - 3 instances
    • bd reopen - 1 instance

    All 12 instances fixed with proper JSON unmarshaling.

[0.21.4] - 2025-11-02

Added

  • New Commands:

    • bd status - Database overview command showing issue counts and stats
    • bd comment - Convenient alias for bd comments add
    • bd daemons restart - Restart specific daemon without manual kill/start
    • --json flag for bd stale command
  • Protected Branch Workflow:

    • BEADS_DIR environment variable for custom database location
    • sync.branch configuration for protected branch workflows
    • Git worktree management with sparse checkout for sync branches
      • Only checks out .beads/ in worktrees, minimal disk usage
      • Only used when sync.branch is configured, not for default users
    • Comprehensive protected branch documentation
  • Migration & Validation:

    • Migration inspection tools for AI agents
    • Conflict marker detection in bd import and bd validate
    • Git hooks health check in bd doctor
    • External reference (external_ref) UNIQUE constraint and validation
    • external_ref now primary matching key for import updates

Fixed

  • Critical Fixes:

    • Daemon corruption from git conflicts
    • MCP set_context hangs with stdio transport (GH #153)
    • Double-release race condition in importInProgress flag
    • Critical daemon race condition causing stale exports
  • Configuration & Migration:

    • bd migrate now detects and sets missing issue_prefix config
    • Config system refactored (renamed config.json → metadata.json)
    • Config version update in migrate command
  • Daemon & RPC:

    • bd doctor --json flag not working
    • bd import now flushes JSONL immediately for daemon visibility
    • Panic recovery in RPC handleConnection
    • Daemon auto-upgrades database version instead of exiting
  • Windows Compatibility:

    • Windows test failures (path handling, bd binary references)
    • Windows CI: forward slashes in git hook shell scripts
    • TestMetricsSnapshot/uptime flakiness on Windows
  • Code Quality:

    • All golangci-lint errors fixed - linter now passes cleanly
    • All gosec, misspell, and unparam linter warnings resolved
    • Tightened file permissions and added security exclusions

Changed

  • Daemon automatically upgrades database schema version instead of exiting
  • Git worktree management for sync branches uses sparse checkout (.beads/ only)
  • Improved test isolation and performance optimization

[0.21.2] - 2025-11-01

Changed

  • Homebrew formula now auto-published in main repo via GoReleaser
  • Deprecated separate homebrew-beads tap repository

[0.21.1] - 2025-10-31

Changed

  • Version bump for consistency across CLI, MCP server, and plugin

[0.20.1] - 2025-10-31

Breaking Changes

  • Hash-Based IDs Now Default: Sequential IDs (bd-1, bd-2) replaced with hash-based IDs (bd-a1b2, bd-f14c)
    • 4-character hashes for 0-500 issues
    • 5-character hashes for 500-1,500 issues
    • 6-character hashes for 1,500-10,000 issues
    • Progressive length extension prevents collisions with birthday paradox math
    • Migration required: Run bd migrate to upgrade schema (removes issue_counters table)
    • Existing databases continue working - migration is opt-in
    • Dramatically reduces merge conflicts in multi-worker/multi-branch workflows
    • Eliminates ID collision issues when multiple agents create issues concurrently

Removed

  • Sequential ID Generation: Removed SyncAllCounters(), AllocateNextID(), and collision remapping logic (bd-c7af, bd-8e05, bd-4c74)
    • Hash IDs handle collisions by extending hash length, not remapping
    • issue_counters table removed from schema
    • --resolve-collisions flag removed from import (no longer needed)
    • 400+ lines of obsolete collision handling code removed

Changed

  • Collision Handling: Automatic hash extension on collision instead of ID remapping
    • Much simpler and more reliable than sequential remapping
    • No cross-branch coordination needed
    • Birthday paradox ensures extremely low collision rates

Migration Notes

For users upgrading from 0.20.0 or earlier:

  1. Run bd migrate to detect and upgrade old database schemas
  2. Database continues to work without migration, but you'll see warnings
  3. Hash IDs provide better multi-worker reliability at the cost of non-numeric IDs
  4. Old sequential IDs like bd-152 become hash IDs like bd-f14c

See README.md for hash ID format details and birthday paradox collision analysis.

[0.20.0] - 2025-10-30

Added

  • Hash-Based IDs: New collision-resistant ID system (bd-168, bd-166, bd-167)
    • 6-character hash IDs with progressive 7/8-char fallback on collision
    • Opt-in via .beads/config.toml with id_mode = "hash"
    • Migration tool: bd migrate --to-hash-ids for existing databases
    • Prefix-optional ID parsing (e.g., bd-abc123 or just abc123)
    • Hierarchical child ID generation for discovered-from relationships
  • Substring ID Matching: All bd commands now support partial ID matching
    • bd show abc matches any ID containing "abc" (e.g., bd-abc123)
    • Ambiguous matches show helpful error with all candidates
  • Daemon Registry: Multi-daemon management for multiple workspaces
    • bd daemons list shows all running daemons across workspaces
    • bd daemons health detects version mismatches and stale sockets
    • bd daemons logs <workspace> for per-daemon log viewing
    • bd daemons killall to restart all daemons after upgrades

Fixed

  • Test Stability: Deprecated sequence-ID collision tests
    • Kept TestFiveCloneCollision for hash-ID multi-clone testing
    • Fixed TestTwoCloneCollision to use merge instead of rebase
  • Linting: golangci-lint v2.5.0 compatibility
    • Added version: 2 field to .golangci.yml
    • Renamed exclude to exclude-patterns for v3 format

Changed

  • Multiple bd Detection: Warning when multiple bd binaries in PATH (PR #182)
    • Prevents confusion from version conflicts
    • Shows locations of all bd binaries found

[0.17.7] - 2025-10-26

Fixed

  • Test Isolation: Export test failures due to hash caching between subtests
    • Added ClearAllExportHashes() method to SQLiteStorage for test isolation
    • Export tests now properly reset state between subtests
    • Fixes intermittent test failures when running full test suite

[0.17.2] - 2025-10-25

Added

  • Configurable Sort Policy: bd ready --sort flag for work queue ordering
    • hybrid (default): Priority-weighted by staleness
    • priority: Strict priority ordering for autonomous systems
    • oldest: Pure FIFO for long-tail work
  • Release Automation: New scripts for streamlined releases
    • scripts/release.sh: Full automated release (version bump, tests, tag, Homebrew, install)
    • scripts/update-homebrew.sh: Automated Homebrew formula updates

Fixed

  • Critical: Database reinitialization test re-landed with CI fixes
    • Windows: Fixed git path handling (forward slash normalization)
    • Nix: Skip test when git unavailable
    • JSON: Increased scanner buffer to 64MB for large issues
  • Bug: Stale daemon socket detection
    • MCP server now health-checks cached connections before use
    • Auto-reconnect with exponential backoff on stale sockets
    • Handles daemon restarts/upgrades gracefully
  • Linting: Fixed all errcheck warnings in production code
    • Proper error handling for database resources and transactions
    • Graceful EOF handling in interactive input
  • Linting: Fixed revive style issues
    • Removed unused parameters, renamed builtin shadowing
  • Linting: Fixed goconst warnings

[0.17.0] - 2025-10-24

Added

  • Git Hooks: Automatic installation prompt during bd init
    • Eliminates race condition between auto-flush and git commits
    • Pre-commit hook: Flushes pending changes immediately before commit
    • Post-merge hook: Imports updated JSONL after pull/merge
    • Optional installation with Y/n prompt (defaults to yes)
    • See examples/git-hooks/README.md for details
  • Duplicate Detection: New bd duplicates command for finding and merging duplicate issues (bd-119, bd-203)
    • Automated duplicate detection with content-based matching
    • --auto-merge flag for batch merging duplicates
    • --dry-run mode to preview merges before execution
    • Helps maintain database cleanliness after imports
  • External Reference Import: Smart import matching using external_ref field (bd-66-74, GH #142)
    • Issues with external_ref match by reference first, not content
    • Enables hybrid workflows with Jira, GitHub, Linear
    • Updates existing issues instead of creating duplicates
    • Database index on external_ref for fast lookups
  • Multi-Database Warning: Detect and warn about nested beads databases
    • Prevents accidental creation of multiple databases in hierarchy
    • Helps users avoid confusion about which database is active

Fixed

  • Critical: Database reinitialization data loss bug (bd-130, DATABASE_REINIT_BUG.md)
    • Fixed bug where removing .beads/ and running bd init would lose git-tracked issues
    • Now correctly imports from JSONL during initialization
    • Added comprehensive tests (later reverted due to CI issues on Windows/Nix)
  • Critical: Foreign key constraint regression (bd-62, GH #144)
    • Pinned modernc.org/sqlite to v1.38.2 to avoid FK violations
    • Prevents database corruption from upstream regression
  • Critical: Install script safety (GH #143 by @marcodelpin)
    • Prevents shell corruption from directory deletion during install
    • Restored proper error codes for safer installation
  • Bug: Daemon auto-start reliability
    • Daemon now responsive immediately, runs initial sync in background
    • Fixes timeout issues when git pull is slow
    • Skip daemon-running check for forked child process
  • Bug: Dependency timestamp churn during auto-import (bd-45, bd-137)
    • Auto-import no longer updates timestamps on unchanged dependencies
    • Eliminates perpetually dirty JSONL from metadata changes
  • Bug: Import reporting accuracy (bd-49, bd-88)
    • bd import now correctly reports "X updated, Y unchanged" instead of "0 updated"
    • Better visibility into import operation results
  • Bug: Memory database handling
    • Fixed :memory: database connection with shared cache mode
    • Proper URL construction for in-memory testing

Changed

  • Removed: Deprecated bd repos command
    • Global daemon architecture removed in favor of per-project daemons
    • Eliminated cross-project database confusion
  • Documentation: Major reorganization and improvements
    • Condensed README, created specialized docs (QUICKSTART.md, ADVANCED.md, etc.)
    • Enhanced "Why not GitHub Issues?" FAQ section
    • Added Beadster to Community & Ecosystem section

Performance

  • Test coverage improvements: 46.0% → 57.7% (+11.7%)
    • Added tests for RPC, storage, cmd/bd helpers
    • New test files: coverage_test.go, helpers_test.go, epics_test.go

Community

  • Community contribution by @marcodelpin (install script safety fixes)
  • Dependabot integration for automated dependency updates

[0.16.0] - 2025-10-23

Added

  • Automated Releases: GoReleaser workflow for cross-platform binaries
    • Automatic GitHub releases on version tags
    • Linux, macOS, Windows binaries for amd64 and arm64
    • Checksums and changelog generation included
  • PyPI Automation: Automated MCP server publishing to PyPI
    • GitHub Actions workflow publishes beads-mcp on version tags
    • Eliminates manual PyPI upload step
  • Sandbox Mode: --sandbox flag for Claude Code integration
    • Isolated environment for AI agent experimentation
    • Prevents production database modifications during testing

Fixed

  • Critical: Idempotent import timestamp churn
    • Prevents timestamp updates when issue content unchanged
    • Reduces JSONL churn and git noise from repeated imports
  • Bug: Windows CI test failures (bd-60, bd-99)
    • Fixed path separator issues and file handling on Windows
    • Skipped flaky tests to stabilize CI

Changed

  • Configuration Migration: Unified config management with Viper (bd-40-44, bd-78)
    • Migrated from manual env var handling to Viper
    • Bound all global flags to Viper for consistency
    • Kept bd config independent from Viper for modularity
    • Added comprehensive configuration tests
  • Documentation Refactor: Improved documentation structure
    • Condensed main README
    • Created specialized guides (QUICKSTART.md, CONFIG.md, etc.)
    • Enhanced FAQ and community sections

Testing

  • Hardened issueDataChanged with type-safe comparisons
  • Improved test isolation and reliability

[0.15.0] - 2025-10-23

Added

  • Configuration System: New bd config command for managing configuration (GH #115)
    • Environment variable definitions with validation
    • Configuration file support (TOML/YAML/JSON)
    • Get/set/list/unset commands for user-friendly management
    • Validation and type checking for config values
    • Documentation in CONFIG.md

Fixed

  • MCP Server: Smart routing for lifecycle status changes in update tool (GH #123)
    • update(status="closed") now routes to close() tool to respect approval workflows
    • update(status="open") now routes to reopen() tool to respect approval workflows
    • Prevents bypass of Claude Code approval settings for lifecycle events
    • bd CLI remains unopinionated; routing happens only in MCP layer
    • Users can now safely auto-approve benign updates (priority, notes) without exposing closure bypass

[0.14.0] - 2025-10-22

Added

  • Lifecycle Safety Documentation: Complete documentation for UnderlyingDB() usage
    • Added tracking guidelines for database lifecycle safety
    • Documented transaction management best practices
    • Prevents UAF (use-after-free) bugs in extensions

Fixed

  • Critical: Git worktree detection and warnings
    • Added automatic detection when running in git worktrees
    • Displays prominent warning if daemon mode is active in worktree
    • Prevents daemon from committing/pushing to wrong branch
    • Documents --no-daemon flag as solution for worktree users
  • Critical: Multiple daemon race condition
    • Implemented file locking (daemon.lock) to prevent multiple daemons per repository
    • Uses flock on Unix, LockFileEx on Windows for process-level exclusivity
    • Lock held for daemon lifetime, automatically released on exit
    • Eliminates race conditions in concurrent daemon start attempts
    • Backward compatible: Falls back to PID check for pre-lock daemons during upgrades
  • Bug: daemon.lock tracked in git
    • Removed daemon.lock from git tracking
    • Added to .gitignore to prevent future commits
  • Bug: Regression in Nix Flake (#110)
    • Fixed flake build issues
    • Restored working Nix development environment

Changed

  • UnderlyingDB() deprecated for most use cases
    • New UnderlyingConn(ctx) provides safer scoped access
    • Reduced risk of UAF bugs in database extensions
    • Updated EXTENDING.md with migration guide

Documentation

  • Complete release process documentation in RELEASING.md
  • Enhanced EXTENDING.md with lifecycle safety patterns
  • Added UnderlyingDB() tracking guidelines

[0.11.0] - 2025-10-22

Added

  • Issue Merging: New bd merge command for consolidating duplicate issues (bd-7, bd-11-17)
    • Merge multiple source issues into a single target issue
    • Automatically migrates all dependencies and dependents to target
    • Updates text references (bd-X mentions) across all issue fields
    • Closes source issues with "Merged into bd-Y" reason
    • Supports --dry-run for validation without changes
    • Example: bd merge bd-42 bd-43 --into bd-41
  • Multi-ID Operations: Batch operations for increased efficiency (bd-195, #101)
    • bd update: Update multiple issues at once
    • bd show: View multiple issues in single call
    • bd label add/remove: Apply labels to multiple issues
    • bd close: Close multiple issues with one command
    • bd reopen: Reopen multiple issues together
    • Example: bd close bd-1 bd-2 bd-3 --reason "Done"
  • Daemon RPC Improvements: Enhanced sync operations
    • bd sync now works correctly in daemon mode
    • Export operations properly supported via RPC
    • Prevents database access conflicts during sync
  • Acceptance Criteria Alias: Added --acceptance-criteria flag (bd-228, #102)
    • Backward-compatible alias for --acceptance in bd update
    • Improves clarity and matches field name

Fixed

  • Critical: Test isolation and database pollution (bd-1, bd-15, bd-19, bd-52)
    • Comprehensive test isolation ensuring tests never pollute production database
    • Fixed stress test issues writing 1000+ test issues to production
    • Quarantined RPC benchmarks to prevent pollution
    • Added database isolation canary tests
  • Critical: Daemon cache staleness
    • Daemon now detects external database modifications via mtime check
    • Prevents serving stale data after external bd import, rm bd.db, etc.
    • Cache automatically invalidates when DB file changes
  • Critical: Counter desync after deletions
    • Issue counters now sync correctly after bulk deletions
    • Prevents ID gaps and counter drift
  • Critical: Labels and dependencies not persisted in daemon mode (#101)
    • Fixed label operations failing silently in daemon mode
    • Fixed dependency operations not saving in daemon mode
    • Both now correctly propagate through RPC layer
  • Daemon sync support: bd sync command now works in daemon mode
    • Previously crashed with nil pointer when daemon running
    • Export operations now properly routed through RPC
  • Acceptance flag normalization: Unified --acceptance flag behavior (bd-228, #102)
    • Added --acceptance-criteria as clearer alias
    • Both flags work identically for backward compatibility
  • Auto-import Git conflicts: Better detection of merge conflicts
    • Auto-import detects and warns about unresolved Git merge conflicts
    • Prevents importing corrupted JSONL with conflict markers
    • Clear instructions for resolving conflicts

Changed

  • BREAKING: Removed global daemon socket fallback
    • Each project now must use its own local daemon (.beads/bd.sock)
    • Prevents cross-project daemon connections and database pollution
    • Migration: Stop any global daemon and restart with bd daemon in each project
    • Warning displayed if old global socket (~/.beads/bd.sock) is found
  • Database cleanup: Project database cleaned from 1000+ to 55 issues
    • Removed accumulated test pollution from stress testing
    • Renumbered issues for clean ID space (bd-1 through bd-55)
    • Better test isolation prevents future pollution

Deprecated

  • Global daemon socket support (see BREAKING change above)

[0.10.0] - 2025-10-20

Added

  • Agent Onboarding: New bd onboard command for agent-first documentation
    • Outputs structured instructions for agents to integrate bd into documentation
    • Bootstrap workflow: Add 'BEFORE ANYTHING ELSE: run bd onboard' to AGENTS.md
    • Agent adapts instructions to existing project structure
    • More agentic approach vs. direct string replacement
    • Updates README with new bootstrap workflow

[0.9.11] - 2025-10-20

Added

  • Labels Documentation: Comprehensive LABELS.md guide (bd-159, bd-163)
    • Complete label system documentation with workflows and best practices
    • Common label patterns (components, domains, size, quality gates, releases)
    • Advanced filtering techniques and integration examples
    • Added Labels section to README with quick reference

Fixed

  • Critical: MCP server crashes on None/null responses (bd-172, fixes #79)
    • Added null safety checks in list_issues(), ready(), and stats() methods
    • Returns empty arrays/dicts instead of crashing on None responses
    • Prevents TypeError when daemon returns empty results

[0.9.10] - 2025-10-18

Added

  • Label Filtering: Enhanced bd list command with label-based filtering
    • --label (or -l): Filter by multiple labels with AND semantics (must have ALL)
    • --label-any: Filter by multiple labels with OR semantics (must have AT LEAST ONE)
    • Examples:
      • bd list --label backend,urgent: Issues with both 'backend' AND 'urgent'
      • bd list --label-any frontend,backend: Issues with either 'frontend' OR 'backend'
    • Works in both daemon and direct modes
    • Includes comprehensive test coverage
  • Log Rotation: Automatic daemon log rotation with configurable limits
    • Prevents unbounded log file growth for long-running daemons
    • Configurable via environment variables: BEADS_DAEMON_LOG_MAX_SIZE, BEADS_DAEMON_LOG_MAX_BACKUPS, BEADS_DAEMON_LOG_MAX_AGE, BEADS_DAEMON_LOG_COMPRESS
    • Optional compression of rotated logs
    • Defaults: 50MB max size, 7 backups, 30 day retention, compression enabled
  • Batch Deletion: Enhanced bd delete command with batch operations
    • Delete multiple issues at once: bd delete bd-1 bd-2 bd-3 --force
    • Read from file: bd delete --from-file deletions.txt --force
    • Dry-run mode: --dry-run to preview deletions before execution
    • Cascade mode: --cascade to recursively delete all dependents
    • Force mode: --force to orphan dependents instead of failing
    • Atomic transactions: all deletions succeed or none do
    • Comprehensive statistics: tracks deleted issues, dependencies, labels, and events

Fixed

  • Critical: bd list --status all showing 0 issues
    • Status filter now treats "all" as special value meaning "show all statuses"
    • Previously treated "all" as literal status value, matching no issues

[0.9.9] - 2025-10-17

Added

  • Daemon RPC Architecture: Production-ready RPC protocol for client-daemon communication (bd-110, bd-111, bd-112, bd-114, bd-117)
    • Unix socket-based RPC enables faster command execution via long-lived daemon process
    • Automatic client detection with graceful fallback to direct mode
    • Serializes SQLite writes and batches git operations to prevent concurrent access issues
    • Resolves database corruption, git lock contention, and ID counter conflicts with multiple agents
    • Comprehensive integration tests and stress testing with 4+ concurrent agents
  • Issue Deletion: bd delete command for removing issues with comprehensive cleanup
    • Safely removes issues from database and JSONL export
    • Cleans up dependencies and references to deleted issues
    • Works correctly with git-based workflows
  • Issue Restoration: bd restore command for recovering compacted/deleted issues
    • Restores issues from git history when needed
    • Preserves references and dependency relationships
  • Prefix Renaming: bd rename-prefix command for batch ID prefix changes
    • Updates all issue IDs and text references throughout the database
    • Useful for project rebranding or namespace changes
  • Comprehensive Testing: Added scripttest-based integration tests (#59)
    • End-to-end coverage for CLI workflows
    • Tests for init command edge cases

Fixed

  • Critical: Metadata errors causing crashes on first import
    • Auto-import now treats missing metadata as first import instead of failing
    • Eliminates initialization errors in fresh repositories
  • Critical: N+1 query pattern in auto-import
    • Replaced per-issue queries with batch fetching
    • Dramatically improves performance for large imports
  • Critical: Duplicate issue imports
    • Added deduplication logic to prevent importing same issue multiple times
    • Maintains data integrity during repeated imports
  • Bug: Auto-flush missing after renumber/rename-prefix
    • Commands now properly export to JSONL after completion
    • Ensures git sees latest changes immediately
  • Bug: Renumber ID collision with UUID temp IDs
    • Uses proper UUID-based temporary IDs to prevent conflicts during renumbering
    • ID counter now correctly syncs after renumbering operations
  • Bug: Collision resolution dependency handling
    • Uses unchecked dependency addition during collision remapping
    • Prevents spurious cycle detection errors
  • Bug: macOS crashes documented (closes #3, bd-87)
    • Added CGO_ENABLED=1 workaround documentation for macOS builds

Changed

  • CLI commands now prefer RPC when daemon is running
    • Improved error reporting and diagnostics for RPC failures
    • More consistent exit codes and status messages
  • Internal command architecture refactored for RPC client/server sharing
    • Reduced code duplication between direct and daemon modes
    • Improved reliability of background operations
  • Ready work sort order flipped to show oldest issues first
    • Helps prioritize long-standing work items

Performance

  • Faster command execution through RPC-backed daemon (up to 10x improvement)
  • N+1 query elimination in list/show operations
  • Reduced write amplification from improved auto-flush behavior
  • Cycle detection performance benchmarks added

Testing

  • Integration tests for daemon RPC request/response flows
  • End-to-end coverage for delete/restore lifecycles
  • Regression tests for metadata handling, auto-flush, ID counter sync
  • Comprehensive tests for collision detection in auto-import

Documentation

  • Release process documentation added (RELEASING.md)
  • Multiple workstreams warning banner for development coordination

[0.9.8] - 2025-10-16

Added

  • Background Daemon Mode: bd daemon command for continuous auto-sync (#bd-386)
    • Watches for changes and automatically exports to JSONL
    • Monitors git repository for incoming changes and auto-imports
    • Production-ready with graceful shutdown, PID file management, and signal handling
    • Eliminates manual export/import in active development workflows
  • Git Synchronization: bd sync command for automated git workflows (#bd-378)
    • One-command sync: stage, commit, pull, push JSONL changes
    • Automatic merge conflict resolution with collision remapping
    • Status reporting shows sync progress and any issues
    • Ideal for distributed teams and CI/CD integration
  • Issue Compaction: bd compact command to summarize old closed issues (bd-254-264)
    • AI-powered summarization using Claude Haiku
    • Reduces database size while preserving essential information
    • Configurable thresholds for age, dependencies, and references
    • Compaction status visible in bd show output
  • Label and Title Filtering: Enhanced bd list command (#45, bd-269)
    • Filter by labels: bd list --label bug,critical
    • Filter by title: bd list --title "auth"
    • Combine with status/priority filters
  • List Output Formats: bd list --format flag for custom output (PR #46)
    • Format options: default, compact, detailed, json
    • Better integration with scripts and automation tools
  • MCP Reopen Support: Reopen closed issues via MCP server
    • Claude Desktop plugin can now reopen issues
    • Useful for revisiting completed work
  • Cross-Type Cycle Prevention: Dependency cycles detected across all types
    • Prevents A→B→A cycles even when mixing blocks, related, etc.
    • Semantic validation for parent-child direction
    • Diagnostic warnings when cycles detected

Fixed

  • Critical: Auto-import collision skipping bug (bd-393, bd-228)
    • Import would silently skip collisions instead of remapping
    • Could cause data loss when merging branches
    • Now correctly applies collision resolution with remapping
  • Critical: Transaction state corruption
    • Nested transactions could corrupt database state
    • Fixed with proper transaction boundary handling
  • Critical: Concurrent temp file collisions (bd-306, bd-373)
    • Multiple bd processes would collide on shared .tmp filename
    • Now uses PID suffix for temp files: .beads/issues.jsonl.tmp.12345
  • Critical: Circular dependency detection gaps
    • Some cycle patterns were missed by detection algorithm
    • Enhanced with comprehensive cycle prevention
  • Bug: False positive merge conflict detection (bd-313, bd-270)
    • Auto-import would detect conflicts when none existed
    • Fixed with improved Git conflict marker detection
  • Bug: Import timeout with large issue sets
    • 200+ issue imports would timeout
    • Optimized import performance
  • Bug: Collision resolver missing ID counter sync
    • After remapping, ID counters weren't updated
    • Could cause duplicate IDs in subsequent creates
  • Bug: NULL handling in statistics for empty databases (PR #37)
    • bd stats would crash on newly initialized databases
    • Fixed NULL value handling in GetStatistics

Changed

  • Compaction removes snapshot/restore (simplified to permanent decay)
  • Export file writing refactored to avoid Windows Defender false positives (PR #31)
  • Error handling improved in auto-import and fallback paths (PR #47)
  • Reduced cyclomatic complexity in main.go (PR #48)
  • MCP integration tests fixed and linting cleaned up (PR #40)

Performance

  • Cycle detection benchmarks added
  • Import optimization for large issue sets
  • Export uses PID-based temp files to avoid lock contention

Community

  • Merged PR #31: Windows Defender mitigation for export
  • Merged PR #37: Fix NULL handling in statistics
  • Merged PR #38: Nix flake for declarative builds
  • Merged PR #40: MCP integration test fixes
  • Merged PR #45: Label and title filtering for bd list
  • Merged PR #46: Add --format flag to bd list
  • Merged PR #47: Error handling consistency
  • Merged PR #48: Cyclomatic complexity reduction

[0.9.2] - 2025-10-14

Added

  • One-Command Dependency Creation: --deps flag for bd create (#18)
    • Create issues with dependencies in a single command
    • Format: --deps type:id or just --deps id (defaults to blocks)
    • Multiple dependencies: --deps discovered-from:bd-20,blocks:bd-15
    • Whitespace-tolerant parsing
    • Particularly useful for AI agents creating discovered-from issues
  • External Reference Tracking: external_ref field for linking to external trackers
    • Link bd issues to GitHub, Jira, Linear, etc.
    • Example: bd create "Issue" --external-ref gh-42
    • bd update supports updating external references
    • Tracked in JSONL for git portability
  • Metadata Storage: Internal metadata table for system state
    • Stores import hash for idempotent auto-import
    • Enables future extensibility for system preferences
    • Auto-migrates existing databases
  • Windows Support: Complete Windows 11 build instructions (#10)
    • Tested with mingw-w64
    • Full CGo support documented
    • PATH setup instructions
  • Go Extension Example: Complete working example of database extensions (#15)
    • Demonstrates custom table creation
    • Shows cross-layer queries joining with issues
    • Includes test suite and documentation
  • Issue Type Display: bd list now shows issue type in output (#17)
    • Better visibility: bd-1 [P1] [bug] open
    • Helps distinguish bugs from features at a glance

Fixed

  • Critical: Dependency tree deduplication for diamond dependencies (bd-85, #1)
    • Fixed infinite recursion in complex dependency graphs
    • Prevents duplicate nodes at same level
    • Handles multiple blockers correctly
  • Critical: Hash-based auto-import replaces mtime comparison
    • Git pull updates mtime but may not change content
    • Now uses SHA256 hash to detect actual changes
    • Prevents unnecessary imports after git operations
  • Critical: Parallel issue creation race condition (PR #8, bd-66)
    • Multiple processes could generate same ID
    • Replaced in-memory counter with atomic database counter
    • Syncs counters after import to prevent collisions
    • Comprehensive test coverage

Changed

  • Auto-import now uses content hash instead of modification time
  • Dependency tree visualization improved for complex graphs
  • Better error messages for dependency operations

Community

  • Merged PR #8: Parallel issue creation fix
  • Merged PR #10: Windows build instructions
  • Merged PR #12: Fix quickstart EXTENDING.md link
  • Merged PR #14: Better enable Go extensions
  • Merged PR #15: Complete Go extension example
  • Merged PR #17: Show issue type in list output

[0.9.1] - 2025-10-14

Added

  • Incremental JSONL Export: Major performance optimization
    • Dirty issue tracking system to only export changed issues
    • Auto-flush with 5-second debounce after CRUD operations
    • Automatic import when JSONL is newer than database
    • --no-auto-flush and --no-auto-import flags for manual control
    • Comprehensive test coverage for auto-flush/import
  • ID Space Partitioning: Explicit ID assignment for parallel workers
    • bd create --id worker1-100 for controlling ID allocation
    • Enables multiple agents to work without conflicts
    • Documented in CLAUDE.md for agent workflows
  • Auto-Migration System: Seamless database schema upgrades
    • Automatically adds dirty_issues table to existing databases
    • Silent migration on first access after upgrade
    • No manual intervention required

Fixed

  • Critical: Race condition in dirty tracking (TOCTOU bug)
    • Could cause data loss during concurrent operations
    • Fixed by tracking specific exported IDs instead of clearing all
  • Critical: Export with filters cleared all dirty issues
    • Status/priority filters would incorrectly mark non-matching issues as clean
    • Now only clears issues that were actually exported
  • Bug: Malformed ID detection never worked
    • SQLite CAST returns 0 for invalid strings, not NULL
    • Now correctly detects non-numeric ID suffixes like "bd-abc"
    • No false positives on legitimate zero-prefixed IDs
  • Bug: Inconsistent dependency dirty marking
    • Duplicated 20+ lines of code in AddDependency/RemoveDependency
    • Refactored to use shared markIssuesDirtyTx() helper
  • Fixed unchecked error in import.go when unmarshaling JSON
  • Fixed unchecked error returns in test cleanup code
  • Removed duplicate test code in dependencies_test.go
  • Fixed Go version in go.mod (was incorrectly set to 1.25.2)

Changed

  • Export now tracks which specific issues were exported
  • ClearDirtyIssuesByID() added (ClearDirtyIssues() deprecated with race warning)
  • Dependency operations use shared dirty-marking helper (DRY)

Performance

  • Incremental export: Only writes changed issues (vs full export)
  • Regex caching in ID replacement: 1.9x performance improvement
  • Automatic debounced flush prevents excessive I/O

[0.9.0] - 2025-10-12

Added

  • Collision Resolution System: Automatic ID remapping for import collisions
    • Reference scoring algorithm to minimize updates during remapping
    • Word-boundary regex matching to prevent false replacements
    • Automatic updating of text references and dependencies
    • --resolve-collisions flag for safe branch merging
    • --dry-run flag to preview collision detection
  • Export/Import with JSONL: Git-friendly text format
    • Dependencies embedded in JSONL for complete portability
    • Idempotent import (exact matches detected)
    • Collision detection (same ID, different content)
  • Ready Work Algorithm: Find issues with no open blockers
    • bd ready command shows unblocked work
    • bd blocked command shows what's waiting
  • Dependency Management: Four dependency types
    • blocks: Hard blocker (affects ready work)
    • related: Soft relationship
    • parent-child: Epic/subtask hierarchy
    • discovered-from: Track issues discovered during work
  • Database Discovery: Auto-find database in project hierarchy
    • Walks up directory tree like git
    • Supports $BEADS_DB environment variable
    • Falls back to ~/.beads/default.db
  • Comprehensive Documentation:
    • README.md with 900+ lines of examples and FAQs
    • CLAUDE.md for AI agent integration patterns
    • SECURITY.md with security policy and best practices
    • TEXT_FORMATS.md analyzing JSONL approach
    • EXTENDING.md for database extension patterns
    • GIT_WORKFLOW.md for git integration
  • Examples: Real-world integration patterns
    • Python agent implementation
    • Bash agent script
    • Git hooks for automatic export/import
    • Branch merge workflow with collision resolution
    • Claude Desktop MCP integration (coming soon)

Changed

  • Switched to JSONL as source of truth (from binary SQLite)
  • SQLite database now acts as ephemeral cache
  • Issue IDs generated with numerical max (not alphabetical)
  • Export sorts issues by ID for consistent git diffs

Security

  • SQL injection protection via allowlisted field names
  • Input validation for all issue fields
  • File path validation for database operations
  • Warnings about not storing secrets in issues

[0.1.0] - Initial Development

Added

  • Core issue tracking (create, update, list, show, close)
  • SQLite storage backend
  • Dependency tracking with cycle detection
  • Label support
  • Event audit trail
  • Full-text search
  • Statistics and reporting
  • bd init for project initialization
  • bd quickstart interactive tutorial

Version History

  • 0.9.8 (2025-10-16): Daemon mode, git sync, compaction, critical bug fixes
  • 0.9.2 (2025-10-14): Community PRs, critical bug fixes, and --deps flag
  • 0.9.1 (2025-10-14): Performance optimization and critical bug fixes
  • 0.9.0 (2025-10-12): Pre-release polish and collision resolution
  • 0.1.0: Initial development version

Upgrade Guide

Upgrading to 0.9.8

No breaking changes. All changes are backward compatible:

  • bd daemon: New optional background service for auto-sync workflows
  • bd sync: New optional git integration command
  • bd compact: New optional command for issue summarization (requires Anthropic API key)
  • --format flag: Optional new feature for bd list
  • Label/title filters: Optional new filters for bd list
  • Bug fixes: All critical fixes are transparent to users

Simply pull the latest version and rebuild:

go install github.com/steveyegge/beads/cmd/bd@latest
# or
git pull && go build -o bd ./cmd/bd

Note: The bd compact command requires an Anthropic API key in $ANTHROPIC_API_KEY environment variable. All other features work without any additional setup.

Upgrading to 0.9.2

No breaking changes. All changes are backward compatible:

  • --deps flag: Optional new feature for bd create
  • external_ref: Optional field, existing issues unaffected
  • Metadata table: Auto-migrates on first use
  • Bug fixes: All critical fixes are transparent to users

Simply pull the latest version and rebuild:

go install github.com/steveyegge/beads/cmd/bd@latest
# or
git pull && go build -o bd ./cmd/bd

Upgrading to 0.9.1

No breaking changes. All changes are backward compatible:

  • Auto-migration: The dirty_issues table is automatically added to existing databases
  • Auto-flush/import: Enabled by default, improves workflow (can disable with flags if needed)
  • ID partitioning: Optional feature, use --id flag only if needed for parallel workers

If you're upgrading from 0.9.0, simply pull the latest version. Your existing database will be automatically migrated on first use.

Upgrading to 0.9.0

No breaking changes. The JSONL export format is backward compatible.

If you have issues in your database:

  1. Run bd export -o .beads/issues.jsonl to create the text file
  2. Commit .beads/issues.jsonl to git
  3. Add .beads/*.db to .gitignore

New collaborators can clone the repo and run:

bd import -i .beads/issues.jsonl

The SQLite database will be automatically populated from the JSONL file.

Future Releases

See open issues tagged with milestone markers for planned features in upcoming releases.

For version 1.0, see: bd dep tree bd-8 (the 1.0 milestone epic)