harness-audit

To audit Claude harness configuration health, run a scan against your ~/.claude/ setup to verify hooks, skills, and agent definitions. This validates frontmatter, tool catalogs, and pipeline-state hygiene to produce a structured verdict of HEALTHY, WARNINGS, or CRITICAL.

Orphan hooks and missing skills in a Claude pipeline occur when configuration changes introduce drift, leaving registered hooks pointing to missing files or referencing non-executable scripts. Scanning the orchestration layer detects these issues by validating JSON and checking that referenced skills exist.

Yes, the harness audit checks for stale agent definitions and invalid JSON by scanning the configuration layer. It validates frontmatter, confirms tool catalogs, and verifies knowledge references to flag non-executable hooks and structural configuration drift.

You should run a configuration health scan on your Claude environment after making any configuration changes to detect drift and prevent broken runs. The audit validates pipeline-state health and checks knowledge references to ensure the orchestration layer remains stable.

Yes, you can validate CLAUDE.md references and pipeline-state hygiene automatically by running an audit on the ~/.claude/ directory. The scan checks knowledge library usage, confirms frontmatter validity, and surfaces configuration issues across hooks and skills.

The difference between HEALTHY, WARNINGS, and CRITICAL harness audit results indicates the severity of configuration issues found. HEALTHY means no issues, WARNINGS flags minor drift like stale agents, and CRITICAL identifies blocking problems like orphan hooks or invalid JSON preventing execution.