| name | triage |
| description | Use after every CI Playwright run to triage failures end-to-end. Runs RCA, hands locator failures to heal, runs the flaky detector over recent runs, files Jira bugs for product defects, returns a one-line summary. |
| allowed-tools | Read, Bash, Edit, Glob, Grep, mcp__atlassian__searchJiraIssuesUsingJql, mcp__atlassian__createJiraIssue |
/triage · the crew
When a Playwright run finishes red on CI (or main), run this skill to triage everything in one shot.
Procedure
- Spawn the
@rca subagent · pass ./test-results/results.json.
- Collect verdicts. If any have
routeTo: heal, spawn @heal per locator failure.
- If
./runs/ has ≥ 3 JSON reports of the same commit, spawn @flaky over the directory.
- Wait for all subagents. Aggregate:
<N> failures classified by RCA · counts per category.<K> locators healed (auto-patched + verified) · <M> escalated to human.<P> flaky tests detected · <Q> quarantined with @flaky.<R> Jira bugs filed.
Output
A markdown summary suitable to paste into a PR comment or Slack thread:
🤖 Triage · ✅ 18/20 specs green · ❌ 2 failures handled
- RCA: 1 product (Jira QA-941 filed) · 1 locator (auto-patched)
- Flaky: 3 detected · 2 quarantined as @flaky
- Heal: 1/1 patched · spec is now green
Stop conditions
- All three sub-pipelines finished.
- Markdown summary printed.
Hard rules
- Do not file the same Jira bug twice · the RCA subagent dedupes via JQL.
- Do not commit any patches · only edit + verify.
