beads/docs/cli-reference/prime.md
matt wilkie 880f34b120 Merge remote-tracking branch 'upstream/main' into pr-4336-fix
Resolves modify/delete conflicts on website/* (Mintlify migration deleted
the tree in main; accept deletion). cmd/bd/prime.go auto-merged cleanly,
re-siting --no-memories on top of #4335's custom-PRIME.md/formatMemoriesForPrime
handling and #4230's doltSync gating without further edits.

Agent-Signature: claude-fable-5-high on behalf of matt wilkie
2026-07-13 21:42:55 -07:00

3.4 KiB

title description
bd prime Output AI-optimized workflow context

{/* AUTO-GENERATED: do not edit manually */}

Generated from bd help --doc prime.

Output essential Beads workflow context in AI-optimized markdown format.

Automatically detects if MCP server is active and adapts output:

  • MCP mode: Brief workflow reminders (~50 tokens)
  • CLI mode: Full command reference (~1-2k tokens)

Designed for Claude Code, Gemini CLI, and Codex SessionStart hooks to prevent agents from forgetting bd workflow after context compaction.

Config options:

  • no-git-ops: When true, outputs stealth mode (no git commands in session close protocol). Set via: bd config set no-git-ops true Useful when you want to control when commits happen manually.

  • agent.profile: Explicit policy profile for git/commit authority wording (conservative | minimal | team-maintainer; default conservative). Set via: bd config set agent.profile team-maintainer Or per-session: BD_AGENT_PROFILE=team-maintainer (env var takes precedence). See docs/getting-started/ide-setup.md#policy-profiles for what each profile means.

    Workflow customization:

    • Place a .beads/PRIME.md file in the local clone or resolved workspace to override the default workflow text. Persistent memories (from bd remember) are still appended so memory injection keeps working under a custom template.
    • Use --export to dump the default content for customization.
    • Use --memories-only for hook contexts that should inject only persistent memories; this returns only the memories section even when a custom PRIME.md is present.
    • Use --no-memories to omit the persistent memories section (useful when the memories section is large and would dominate a context budget). --memories-only takes precedence if both are set.

Memory injection caps: Large memory sets can exceed what a session-start hook host will ingest, and hosts truncate silently. Cap what prime injects with --max-memories N and/or --max-memory-chars N (or the prime.max-memories / prime.max-memory-chars config keys; an explicit flag wins, and an explicit 0 forces unlimited). Caps apply at whole-memory boundaries, at least one memory is always emitted, and a banner ahead of the entries reports how many were elided and how to browse the rest with bd memories. --max-memory-chars caps the total bytes of the injected memory entries; the section header and elision banner are excluded from the budget.

bd prime [flags]

Flags:

      --export                 Output default content (ignores PRIME.md override)
      --full                   Force full CLI output (ignore MCP detection)
      --hook-json              Wrap output in the SessionStart hook JSON envelope (Claude Code, Gemini CLI, Codex)
      --max-memories int       Cap injected persistent memories to N entries (0 = unlimited; falls back to the prime.max-memories config key)
      --max-memory-chars int   Cap the total bytes of injected memory entries, at whole-memory boundaries; section header and banner are not counted (0 = unlimited; falls back to the prime.max-memory-chars config key)
      --mcp                    Force MCP mode (minimal output)
      --memories-only          Output only persistent memories for compact hook contexts
      --no-memories            Omit the persistent memories section (ignored when --memories-only is set, which wins)
      --stealth                Stealth mode (no git operations, flush only)