Skip to content

2.9 Agent action audit trail

2. Validation / PL2-agent-audit-trail

2.9 Agent action audit trail

every agent decision is logged with reasoning, retrievable, and reversible at granular level (catches *quiet drift* — subtle wrong actions humans don't notice for weeks)


Levels

Level 0

No audit trail

Level 1

Commit-level revert only

Level 2

Decision-level reasoning captured; full action history queryable and reversible

Level 3

Patterns extracted from audit log inform future agent prompts; "agents in this codebase tend to over-use approach X" feeds back as guidance. Audit signal also identifies gaps in pre-action gating — actions that slipped past a gate surface here and drive gate refinement


Recipes that advance this criterion