Runs a quality audit on your Claude skills and commands, checking for overlap, staleness, and actionability. You get two modes: Quick Scan diffs only what changed since last run (5-10 min), Full Stocktake reviews everything using sequential subagent batches (20-30 min). It scans both global and project-level skills, checks technical references with WebSearch, then outputs Keep/Improve/Update/Retire/Merge verdicts with specific reasoning. The checklist is deliberately holistic rather than scored, which means you get judgment calls instead of numbers. Most useful when your skills directory has grown organically and you need an honest inventory of what's actually worth keeping. Results cache to JSON so subsequent runs stay fast.
npx skills add https://github.com/affaan-m/everything-claude-code --skill skill-stocktake