flaky-test-investigation-7a63ce— agent skill

Purpose

Flaky tests erode trust in CI. Do not just re-run them and hope for the best — isolate the flake vector, fix it, and prove the fix with a stress loop.

When to Use

• A test fails intermittently in CI but passes locally (or vice versa)
• A test passes alone but fails in the full suite
• A re-run "fixed" a failure and nobody knows why

Inputs

• The flaky test's name/path and the runner command for it
• Recent failing runs, if available, to estimate the failure rate

Workflow

1. Isolate the test. Run the specific failing test by itself. If it passes alone, the flake is likely an order dependency or state leakage from a previous test — run the suite up to and including it to confirm.

1. Stress test. Run the test in a tight loop to establish the failure rate before changing anything:

   for i in {1..100}; do npm test -- -t "My Test" || echo "FAIL on run $i"; done

(Adapt the inner command to the project's runner; some runners have repeat flags built in.)

1. Check the common vectors:

• Time — does the test rely on Date.now() or setTimeout? Mock the clock.
• Async/Promises — asserting before a background task finishes? Ensure proper await or waitFor usage.
• Shared state — reusing database records, global singletons, or mutated variables between runs? Ensure clean teardowns in afterEach.
• Randomness — random IDs or sort orders? Force deterministic seeds or sort orders.

1. Prove the fix. Do not just guess. The fix must be verified by running the stress test loop again and achieving a 100% pass rate.

Output

• The identified flake vector (order/state, time, async, randomness)
• The fix, plus stress-loop evidence (pre-fix failure rate vs post-fix 100% pass)

Verification

• [ ] Test run in isolation to separate order-dependency from intrinsic flake
• [ ] Stress loop run before the fix to establish a baseline failure rate
• [ ] Flake vector named explicitly
• [ ] Stress loop re-run after the fix with a 100% pass rate

Failure Modes

• Re-run and hope — a green re-run proves nothing; the flake is still there.
• Fixing without a baseline — without a pre-fix failure rate, a "fix" cannot be distinguished from luck.
• Quarantining forever — skipping the test removes the signal but keeps the bug.