What problem does it solve? When you record a correction like "don't sort by p-value" in your Claude memory files (lessons.md, axioms.md, or feedback_*.md), the project docs that originally recommended that practice stay stale, and future sessions keep repeating the retracted advice. This skill catches those contradictions automatically. ## Core Features & Use Cases - Event-driven reverse-lint: A PostToolUse hook fires when a memory file is edited, extracts the new negation rule, greps project docs, and surfaces file:line candidate stale claims — never auto-edits. - Dead-branch detection: Finds retired rules whose scoped conditions ("still applies if you are a Y") survive in docs after the condition no longer exists. - Weekly cron audit: A safety-net scan of docs/{research,decisions,findings,runbooks}/ against recent negation rules, including near-duplicate detection at 70% token overlap. - Use Case: You add a lesson "never deploy on Fridays" — the hook immediately flags that docs/runbooks/release.md still recommends Friday deploys, so you can update it before the next session reads it as authoritative. ## Quick Start Ask the assistant to run the doc freshness audit against your project to check whether any docs still contradict your recent lessons and feedback entries.
