What problem does it solve?
It prevents your learned correction rules from staying trapped in corrections/rules.md by auditing them for maturity, redundancy, and contradictions so they can be promoted into CLAUDE.md or hooks.
Core Features & Use Cases
- Improvement-trend visibility: Computes recent effectiveness trends from scorecards.jsonl to show whether the learning system is improving.
- Rule quality classification: Categorizes rules as PROMOTE, REDUNDANT, CONTRADICT, INEFFECTIVE, KEEP, or CROSS-PROJECT based on stability, effectiveness, clustering, and conflicts with existing configuration.
- Automated promotions & cleanup: Automatically appends mature rules into the correct section of CLAUDE.md (or creates Auto-Promoted Rules) and removes promoted/redundant rules from corrections/rules.md.
Quick Start
Run the audit skill to review corrections/rules.md and automatically promote stable rules into your CLAUDE.md.