Audits how well the vercel-plugin injects skills into Claude Code sessions by parsing actual conversation logs. You point it at a project's JSONL logs in ~/.claude/projects, and it extracts every tool call, tests which skills matched using the real hook pipeline, then shows you the gaps. It catches path patterns that should have fired but didn't, commands that slipped through, and skills that matched but got deduped or dropped by the 12KB budget. Also diffs your installed plugin cache against dev to spot staleness. Run this when skill injection feels off on a real project and you need concrete data on what matched, what didn't, and why.
npx skills add https://github.com/vercel-labs/vercel-plugin --skill plugin-audit