What problem does it solve? When a Claude Code plugin injects skills based on tool-call patterns, it is hard to know whether the right skills fire on real projects. This Skill audits vercel-plugin performance by replaying actual conversation logs through the hook matching pipeline, exposing coverage gaps, dedup behavior, and stale plugin caches. ## Core Features & Use Cases - Log-based replay: Extracts tool_use entries (Bash, Read, Write, Edit) from JSONL conversation logs and replays them through the hook's loadSkills/matchSkills functions. - Gap and dedup analysis: Produces a match matrix, dedup timeline, and list of unmatched tool calls so you can identify missing path or Bash patterns and budget/cap drops. - Cache staleness checks: Compares the installed plugin cache against the dev version using installed_plugins.json and content diffs. - Use Case: After noticing a skill never fired during a Next.js session, run the batch-match script against that session's JSONL log to see exactly which tool calls matched, which were deduped, and which patterns need to be added. ## Quick Start Ask the AI to audit vercel-plugin skill injection on your project by analyzing the conversation log at ~/.claude/projects/ for your project directory.
