What problem does it solve? Over time, installed Claude skills and commands accumulate overlap, stale technical references, and low-value content, but manually reviewing dozens of skill files is slow and inconsistent. This Skill automates the audit with a fixed quality checklist, holistic AI judgment, and cached results so repeat reviews only re-evaluate what changed. ## Core Features & Use Cases - Two audit modes: Quick Scan re-evaluates only skills changed since the last run (5–10 min), while Full Stocktake reviews the entire inventory (20–30 min). - Structured verdicts: Each skill receives a Keep, Improve, Update, Retire, or Merge verdict with a self-contained, decision-enabling reason, persisted to a results.json cache with resume support for interrupted runs. - Batched subagent evaluation: Inventories are processed in chunks of roughly 20 skills per general-purpose subagent to keep context manageable, with usage-frequency stats (7d/30d) informing decisions. - Use Case: Run it from a project root to audit both global ~/.claude/skills and project-level .claude/skills, then review a summary table and confirm any retire or merge actions before they happen. ## Quick Start Ask the AI to run /skill-stocktake from your project directory to audit all global and project-level skills and produce a verdict table. ## Quick Start Run /skill-stocktake from your project root to audit all global and project-level skills and receive a verdict summary table.
