What does this skill do?

The Skills Audit systematically evaluates the quality, relevance, and usefulness of all Claude skills and commands in your project or across the board. Using a structured checklist and holistic AI assessment, it issues actionable recommendations—keep, improve, update, remove, or merge—so you can maintain a clean, consistent, and relevant skills library.

Redundancy Detection
Identify obsolete, redundant, or low-quality skills that overlap with each other or with MEMORY.md.
Incremental Quick Scan
Re-evaluate only the skills that have been modified since the last audit in 5–10 minutes.
Comprehensive Review
It performs a 4-phase full stocktake that takes inventory of, evaluates, and consolidates the entire library in 20–30 minutes.
Consolidation with Confirmation
It proposes specific actions for removal, merging, or improvement and requests explicit confirmation before taking action.

Usage examples

🔍 Quick scan
Run /skill-stocktake to view only the skills that have changed this week.
📊 Comprehensive audit
Run /skill-stocktake full to audit the project's entire skill library.
🧹 Removing redundant items
Review my skills and let me know which ones I should remove or merge because they overlap.
📅 Regular maintenance
Check to see if there are any skills with outdated technical references that need to be updated.

Features

Automatic Inventory The scan.sh script lists all skill files, extracts the frontmatter, and records modification dates for 7 and 30 days ago.
Checklist-Based Assessment A subagent evaluates each skill against criteria such as overlap, technical novelty, and frequency of use, returning a JSON verdict with justification.
Batch processing Approximately 20 skills are processed per batch to keep the scope manageable and ensure high-quality evaluations.
Enforceable judgments Five clear verdicts: Keep, Improve, Update, Retire, or Merge into [X], each with a self-explanatory rationale.
Consistent Results Save results.json with the UTC timestamp, the mode used, and the verdicts for future incremental runs.

Frequently asked questions

Quick Scan re-evaluates only the skills that have been modified since the last run (5–10 min). Full Stocktake runs all four phases on the entire library (20–30 min).
No. The skill uses its own scripts (scan.sh, quick-diff.sh) and Claude's general-purpose subagent. It only needs access to the project's file system.
No. Phase 4 proposes actions such as removal or merging, but requires explicit confirmation from the user before any deletion or archiving.
In ~/.claude/skills/skill-stocktake/results.json, with actual UTC timestamps, mode used, batch progress, and persistent verdicts.
Skills Audit — Review and Clean Up Your Claude AI Library

¿Prefieres escuchar el contenido? Genera la narración de audio con un clic.