kud/claude-plugins/ci-diagnose

Step 1 — Resolve the job and build

If the user provided a job name (e.g. /ci-diagnose my-service), use it directly.

If the user provided a job name and build number (e.g. /ci-diagnose my-service 42), use both.

If neither was given, ask: "Which Jenkins job are you looking at, and which build number (or should I check the last build)?"

Step 2 — Fetch the build status

Call get_job_status with the job name to get the last build number and its result (if no build number was specified).

If a specific build number was provided, call get_build_status with the job name and build number.

Step 3 — Fetch the console log

Call get_console_log with the job name and build number.

Limit the log to the most relevant portion — look for lines containing ERROR, FAILED, Exception, exit code, or BUILD FAILURE.

Step 4 — Diagnose the failure

Analyse the console output and identify:

• The root cause — the first real error, not downstream noise
• The affected stage or step
• Any missing dependencies, config issues, or flaky test indicators

Step 5 — Present the diagnosis

Structure your response as:

### CI Diagnosis — <job-name> #<build-number>
 
**Status**: <FAILURE / ABORTED / etc.>
**Root cause**: <one-sentence summary>
 
**Details**:
<explanation of what went wrong and why>
 
**Suggested fix**:
<concrete next step — e.g. "update the dependency", "re-run the flaky test", "check the env var X">

If the build is not failing (status is SUCCESS), say so and skip the diagnosis.

Step 6 — Offer to fetch more context (optional)

If relevant, offer to fetch:

• get_test_results — to see which tests failed
• get_pipeline_stages — to see which pipeline stage broke
• get_build_changes — to correlate with recent commits