audit_skill

Overview

This skill orchestrates repeatable cross-repo backlog review using a 5-phase survey→verify→summarize→act→snapshot workflow. It prevents closing items without verifying the codebase and coordinates parallel, read-only subagents to check briefs against actual code. Mandatory for structured backlog review sessions and requires the bon skill to be loaded first.

How this skill works

First it runs a survey to collect structured data on all open bon items across repositories and flags aged briefs. Next it dispatches read-only subagents to verify each brief against the codebase using file checks, code grep, and git history, then classifies items as DONE, STALE, ACTIVE, BLOCKED, or UNCLEAR. The skill presents a human-readable triage summary and only performs closures or commits after explicit user approval. Finally it re-surveys to snapshot the before/after delta.

When to use it

• Periodic backlog review (fortnightly or monthly)
• After a burst of cross-repo activity
• When open bon item count is growing and needs pruning
• Before starting a focused work period to shed stale commitments
• When you must ensure closures are backed by code verification

Best practices

• Always load the bon skill before running audit
• Confirm scope after the initial survey; audit stops for your approval
• Let subagents be read-only — no mutations until you approve actions
• Close only items classified as DONE or approved STALE after you review evidence
• Run the final snapshot to capture before/after deltas for accountability
• Avoid editing briefs mid-audit; record edits for a follow-up session

Example use cases

• Monthly GTD-style backlog sweep across dozens of repos
• Post-release cleanup to close items that were actually implemented
• Prune stale briefs discovered after a long inactivity period
• Triaging cross-repo dependency items that may be blocked by external work
• Validating that planned tasks were completed before marking them closed