plugin-audit

To audit skill injection coverage from Claude Code logs, you extract tool_use entries from real conversation sessions to validate hook matching and identify coverage gaps. The audit generates a structured report highlighting missing injections and cache staleness for remediation.

Vercel-plugin hook matching is the process of testing skill inputs against loaded skills to determine if the correct skill is injected during a Claude Code session. It uses loadSkills and matchSkills functions to validate that real conversation inputs trigger the expected tool_use responses.

You test hook matching against actual inputs by running the batch-match.mjs script against a Claude Code JSONL log. This workflow processes real conversation data, applies the loadSkills and matchSkills functions, and outputs a structured report detailing matching discrepancies and cache differences.

Yes, you can check vercel-plugin cache freshness by analyzing conversation logs to compare current skill states against cached injections. The audit process extracts tool_use entries and generates a report highlighting cache staleness and deduplication issues for remediation.

Skill injection coverage gaps occur when hook matching fails to trigger the correct tool_use for specific conversation inputs. Running an audit with batch-match.mjs against your JSONL logs identifies these exact gaps, dedup issues, and cache differences in a structured report.

No, the plugin-audit workflow does not require external dependencies. It operates using internal scripts and references to parse JSONL conversation logs and execute the loadSkills and matchSkills workflow steps directly on your real-world Claude Code session data.