Agent-Drafted Flaky-Test Cleanup Plans from JUnit Reports

This workflow adds reasoning to flaky-test triage. When your CI pipeline posts a JUnit report, an agent inspects the failure messages and stack traces, classifies the probable cause (timing, shared state, network, ordering), and drafts a concrete cleanup plan, then files an owner-assigned ClickUp task so the fix starts with a hypothesis instead of a blank page.

Use it when raw flake counts are not enough and triagers waste time re-reading the same stack traces. It suits teams that want first-pass root-cause guesses and a starting fix plan attached to every quarantined test.

1. 1An incoming webhook delivers the JUnit XML from a finished CI run.
2. 2The flow parses the report and isolates non-deterministic failures (retried or intermittently failing cases).
3. 3The agent reads each failure's message and trace, classifies the likely flake category, and proposes a remediation plan.
4. 4It resolves the responsible owner from the repository's CODEOWNERS.
5. 5It creates a ClickUp task per flake with the category, fix plan, and failing-test details assigned to that owner.