Agent Audit Log

Create and maintain a lightweight audit trail for high-value actions.

Core rule

Log only actions that matter for safety, traceability, or later review. Do not turn the audit log into noise.

Default layers

1. Raw fact log (YYYY-MM-DD.jsonl)
2. Date summary (index.json)
3. Target/project index (by-target.json)
4. Risk index (by-risk.json)
5. Human-readable summary (latest.md)
6. Export-safety events (export_safety_check)
7. Open items (open-items.json)
8. Status transition history (open-items-history.json)

Read references as needed

• Read references/schema.md for the log schema and event fields.
• Read references/risk-model.md for how to classify low / medium / high risk.
• Read references/export-safety.md before logging publish/export actions.
• Read references/open-items.md when tracking unresolved risks or follow-up work.
• Read references/examples.md when you need concrete event, export-safety, or open-item examples.

Use scripts as needed

• Use scripts/init_audit.sh to create the basic audit directory and starter files.

Operating rules

• Do not store plaintext secrets in audit logs.
• Prefer concise, human-readable summaries.
• Record target, result, and non-sensitive references.
• Use warn when something needs attention but did not fail.
• Use open items for follow-up risk, not for routine noise.