nightly-check

name	nightly-check
description	User-invoked only. Triage scheduled CI runs: flake vs regression vs config vs infra
disable-model-invocation	true

Check nightly CI runs and triage failures

Review the most recent nightly workflow runs for this repo, diagnose any failures, and action them.

---

Step 1 — Fetch recent scheduled runs

Use the gh CLI to fetch scheduled workflow runs for this repo. Look at the last 7 days of nightly (schedule-triggered) runs.

gh run list --event schedule --limit 30 --json databaseId,displayTitle,status,conclusion,createdAt,headSha,workflowName

Filter to runs from the last 7 days. For each run, note:

• Run ID and workflow name
• Triggered at (date/time)
• Overall status (success / failure / in_progress)

Step 2 — Identify failures

List every failed nightly run. For each failure:

1. Fetch the jobs: gh run view <run-id> --json jobs
2. For each failed job, fetch the full logs: gh run view <run-id> --log-failed
3. Read the logs carefully and identify the root cause.

Categorise each failure as one of:

• Flake — non-deterministic failure (network timeout, race condition, transient infra)
• Regression — a code change broke something
• Config — GitHub Actions workflow config, environment variable, or secret issue
• Infra — underlying infrastructure problem (disk, memory, runner crash)
• Unknown — cannot determine from logs alone

Step 3 — Summarise findings

Print a table:

| Pipeline | Date | Status | Job | Category | Root cause (1 line) |
|----------|------|--------|-----|----------|---------------------|

If all nightly runs are passing, output: "✅ All nightly runs passing — nothing to action." and stop.

Step 4 — For each Regression or Config failure

These are actionable. For each one:

1. Identify the affected files, tests, or config.
2. Read the relevant source files to understand the failure.
3. Propose a concrete fix plan (what to change, where, why).
4. Ask: "Should I create a Jira ticket and pick it up for [failure description]?"
   • If yes → read and follow .agents/skills/capture/SKILL.md to log it, then read and follow .agents/skills/pickup/SKILL.md with the new ticket key to start work.
   • If no → note it and move on.

Step 5 — For each Flake failure

1. Check if the flake has occurred more than twice in the last 7 days.
2. If yes → it is chronic and needs fixing. Propose a fix plan and ask whether to create a ticket.
3. If no → log it as a known flake and move on.

Step 6 — For Infra failures

Flag to the user. These are not fixable via code — they need infra investigation. Note the pipeline number and affected jobs so the user can escalate.

Step 7 — Final report

Summarise what was found and what was actioned:

• Pipelines reviewed
• Failures found (by category)
• Tickets created (if any)
• Recommended next steps