62 lines
1.6 KiB
Markdown
62 lines
1.6 KiB
Markdown
---
|
|
title: "bd audit"
|
|
description: "Record and label agent interactions (append-only JSONL)"
|
|
---
|
|
|
|
{/* AUTO-GENERATED: do not edit manually */}
|
|
|
|
Generated from `bd help --doc audit`.
|
|
|
|
Record explicit agent/tool interaction audit entries in .beads/interactions.jsonl.
|
|
|
|
This optional JSONL sidecar is disabled by default. Enable it with:
|
|
|
|
bd config set audit.enabled true
|
|
|
|
Issue history is always recorded in the database and is visible with
|
|
bd history <id> --events. The JSONL sidecar is for explicit interaction capture:
|
|
- auditing ("why did the agent do that?")
|
|
- dataset generation (SFT/RL fine-tuning)
|
|
|
|
Entries are append-only. Labeling creates a new "label" entry that references a parent entry.
|
|
|
|
```
|
|
bd audit [command]
|
|
```
|
|
|
|
## bd audit label
|
|
|
|
Append a label entry referencing an existing interaction
|
|
|
|
```
|
|
bd audit label <entry-id> [flags]
|
|
```
|
|
|
|
**Flags:**
|
|
|
|
```
|
|
--label string Label value (e.g. "good" or "bad")
|
|
--reason string Reason for label
|
|
```
|
|
|
|
## bd audit record
|
|
|
|
Append an audit interaction entry
|
|
|
|
```
|
|
bd audit record [flags]
|
|
```
|
|
|
|
**Flags:**
|
|
|
|
```
|
|
--error string Error string (llm_call/tool_call)
|
|
--exit-code int Exit code (tool_call) (default -1)
|
|
--issue-id string Related issue id (bd-...)
|
|
--kind string Entry kind (e.g. llm_call, tool_call, label)
|
|
--model string Model name (llm_call)
|
|
--prompt string Prompt text (llm_call)
|
|
--response string Response text (llm_call)
|
|
--stdin Read a JSON object from stdin (must match audit.Entry schema)
|
|
--tool-name string Tool name (tool_call)
|
|
```
|