Flaky Test Diagnostician

Flaky tests erode trust in your test suite. Teams spend 5-10 hours a week rerunning builds, approving over failures, and chasing ghosts — and every flaky build that passes on rerun is a real defect that could have been caught.

This skill systematically isolates a flaky test, reruns it in controlled conditions, and classifies the failure against six known root causes — timing, shared state, network dependencies, selector fragility, environment differences, and order dependency — so you fix the cause, not the symptom.

Phase 1 — THINK

Before running anything, understand the test and its context:

1. What kind of test is it? Unit, integration, E2E (Playwright/Cypress), or component test? Each has different flakiness profiles.
2. What does the test touch? Filesystem, network, database, time, random values, shared fixtures, global state, other tests' leftovers?
3. What's the failure pattern? Does it fail on CI but pass locally? Pass in isolation but fail in the suite? Fail at specific times of day?
4. What's the rerun cost? How long does one run take? N=10 vs N=50 vs N=100 depends on test duration.

Phase 2 — ISOLATE