skill-stocktake

You can audit Claude skills and commands by running an automated batch evaluation that sequentially checks quality using subagents. This Skill supports both Quick Scan for recently changed skills and Full Stocktake for comprehensive reviews.

Use the Quick Scan mode to re-evaluate only Claude skills that have changed since the last run. This mode caches previous results, saving time by skipping unchanged commands during the quality audit.

Yes, bash scripting and Python libraries are required dependencies for executing the sequential subagent batch evaluation. You must have both installed in your environment to run quality audits.

You can audit both project-level skills specific to a single project and global Claude commands across all projects. The Full Stocktake mode provides a complete review of all available skills and commands.

Subagent batch processing maintains manageable context by sequentially evaluating Claude skills in batches rather than simultaneously. Results are cached after each batch evaluation for future reference and analysis.

Full Stocktake performs a comprehensive review of all skills and commands using a checklist and AI holistic judgment, but sequential batch evaluation means the process takes longer for large skill collections. Results caching helps optimize subsequent Quick Scans.