audit-trail-review

name	audit-trail-review
description	Inspect run history, approvals, outbox, interpreter manifests, screenshots, and memory events to reconstruct what the agent actually did.

Audit Trail Review

Purpose

Reconstruct reality from logs and artifacts. This skill helps answer what happened, what was approved, what was sent, what changed, and what remains uncertain.

When To Use

Use when the user asks current status, "did it send", "what changed", "why did this happen", or for compliance/debug audits.

Inputs And Evidence

• Run IDs, approval tokens, outbox IDs, interpreter run IDs, screenshot filenames, commit hashes, and memory events.

Tool Map

• channel_outbox
• python_interpreter_runs
• python_interpreter_run
• screen_captures
• memory_search
• cognitive_state
• status-update-writing

Workflow

1. Identify the time window and artifact types.
2. Inspect relevant native stores/tools.
3. Separate prepared, approved, executed, failed, blocked, and skipped actions.
4. Tie claims to IDs/paths/tool results.
5. Report missing evidence as missing.
6. Recommend cleanup or follow-up if needed.

Native Implementation Boundaries

• Use Humungousaur audit/artifact/status tools.
• Do not import external reference audit-trail plugins.
• Do not infer completion without evidence.

Safety And Approval

• Audit logs may contain sensitive data.
• Redact secrets and private messages.
• Do not delete artifacts unless approved.

Verification

• Cite IDs, paths, commits, or statuses.
• Confirm current worktree/run state.
• Label uncertainty.

Failure Modes

• Relying on memory instead of logs.
• Mixing prepared and sent states.
• Ignoring failed approvals.

References

• Shortlist item: audit-trail-review.