skill-stocktake

Auditing Claude skills involves running an AI-guided stocktake that checks them against a quality checklist to identify gaps. The process evaluates both global and project-level skills, generating a results.json file with per-skill verdicts and rationales.

A skill stocktake is a quality assurance process that audits Claude commands and skills against a predefined checklist. It scans both ~/.claude/skills and project-level .claude/skills directories to generate a machine-readable results.json with evaluated timestamps, batch progress, and per-skill verdicts.

Run /skill-stocktake from the project root to start auditing. A Quick Scan targets recently changed skills for faster checks, while a Full Stocktake performs a complete audit across all global and project-level skills to ensure total visibility.

Yes, you need jq installed as a dependency to run a skill audit. The underlying scripts rely on jq to process and generate the machine-readable results.json output containing per-skill verdicts and batch progress.

The best way to track quality gaps is by generating a centralized results.json file. This machine-readable output records evaluated_at timestamps, batch_progress, per-skill verdicts, and a concise rationale for each decision, ensuring transparent and repeatable assessments.

Yes, the stocktake process scans both ~/.claude/skills for global skills and {cwd}/.claude/skills for project-level skills simultaneously. This ensures full visibility and comprehensive coverage across your entire Claude environment.