| name | diagnose |
| description | Analyze test, CI, benchmark, or eval failures; cluster by root cause; rank by impact and fix effort; produce concrete fix hypotheses. |
Diagnose
Use this when there are multiple failures and the next decision is what to fix first.
Do not patch before clustering; repeated symptoms often share one cause.
Flow
- Locate the raw failure data: logs, reports, CI artifacts, JSON, XML, or benchmark rows.
- Parse all failures and preserve counts.
- Cluster by root cause, not by file or test name alone.
- Rank clusters by user impact, unblock value, confidence, and fix effort.
- For each top cluster, name a testable hypothesis and the exact evidence that supports it.
- Recommend the first fix and the check that should go green.
Output
Return totals, clusters, representative errors, root-cause hypotheses, ranked fix order, and missing data.
Use references/full-reference.md for supported formats and parsing details.
Then consider
converge when the next step is fixing CI.
autopsy when one result is suspicious rather than many failures.