skill-stocktake

You can audit Claude skills for overlap and staleness by running an automated inventory scan that evaluates markdown files using checklist criteria and UTC mtime tracking. It identifies redundant capabilities and outdated content within your skill directories.

The best way to inventory markdown command files is using a YAML-frontmatter-driven identification workflow. This process catalogs global and project-scoped skill directories, applying checklist evaluations to flag quality issues and detect overlapping commands.

Resumable batch processing for quality audits works by caching intermediate progress in a results.json file. If the evaluation is interrupted, the system can resume the audit from the last saved state, preventing redundant scans of already evaluated skills.

Yes, you can use a Quick Scan to review only recently changed skills. This mode compares current file states against cached results using change detection, outputting only the diff of modifications since the last full inventory scan was performed.

Yes, the skill audit requires jq as a dependency to parse JSON results. The evaluation workflow generates JSON outputs for both intermediate caching and final audit reports, which jq processes to extract and format the quality scoring signals.

Usage-aware scoring signals evaluate skill usefulness by incorporating 7-day and 30-day read observations. These usage metrics help judge real-world utility during a Full Stocktake, providing data to determine whether to keep, improve, or retire specific commands.