inspect-metrics

Monitor CLAUDE Code Reviewer metrics by querying the live /metrics endpoint to display current system health, review counts, verdict distribution, and skip reasons across multiple PRs. The dashboard aggregates live endpoint data with a persisted state file for comprehensive operational visibility.

Track PR review activity, errors, and verdicts by viewing metrics that include review counts, verdict distribution, errors by phase, and skip reasons. Drill down by PR status for targeted investigation to surface potential degradation and bottlenecks.

Anomaly detection for code review observability works by applying health signaling to surface potential performance degradation and bottlenecks. It aggregates live endpoint metrics and persisted state data to identify operational performance signals requiring investigation.

Yes, you need a running CLAUDE Code Reviewer instance with a valid config.yaml, plus access to the live metrics endpoint and data/state.json. These components provide the real-time health data and persisted state required for dashboard generation.

Yes, you can retrieve review metrics from data/state.json as a fallback for offline or crash-recovery scenarios. The skill derives fallback metrics from the persisted state file when the live endpoint is unavailable, ensuring historical context and continuity.

The best way to investigate PR review errors by phase and skip reasons is using the dashboard's quick drill-down capabilities. Filter by PR status, verdict distribution, error phases, and skip reasons for targeted investigation of operational bottlenecks.