AI coding

Audit logging for agent actions

• A Task Log for Agent Work: What Was Asked and What Shipped
• Auditing Human Decisions in the Task Log
• A Fallback Evidence Bundle When the Link Dies
• A Tamper Log Schema: Fields, Joins and Alerts
• A Worktree Naming Scheme That Keeps Traces Aligned
• Agent Commit Messages as a Search Key for Later Audits
• Approval Logs That Join Proposals to Applied Rows
• Audit Logs for Agent Actions: What to Record and How to Query
• Audit Logs for Agent Merge Actions: What to Record and Query
• Auditing Agent Database Writes: Triggers, Logs and Who Did What
• Detecting Reads Outside the Canonical Tree
• Force-Push Limits and the Bypass Audit Log
• Logging Disagreements: A Schema for Conflicting Agent Outputs
• Logging Every Policy Decision: Fields to Reconstruct a Refusal
• Logging the Tool Name a Model Chose and the One It Was Allowed
• Naming Agent Sessions: A Convention for Logs and Traces
• Pinning a Run ID to Every Artifact a Reviewer Opens
• Postmortems for Agent Failures: Write the Runbook First
• Reading a Refusal Log: What an Agent Tried to Open
• Reading an Agent Sandbox's Keystore Log After a Signing Anomaly
• Reading an Override Log for Patterns: Queries and Thresholds
• Recording Human Approvals in an Agent's Audit Trail
• Recording the Decision a Model Made on Each Error: A Log Schema
• Rotating Trace Storage Without Breaking Old Review Links
• What a Wrapper Logs When an Agent Calls It: Fields for Audit
• What to Log About an Agent Run to Reconstruct It
• When a Rule Fires Wrong: Subject vs Action Match
• When to Count a Retry as a New Task in Audit Logs