What problem does it solve? Tests that pass sometimes and fail sometimes — locally vs CI, alone vs in suite — hide real causes behind retry-until-green band-aids, tripling CI cost and occasionally masking genuine races in production code. This Skill provides a structured reproduction loop that identifies the exact root cause instead of papering over it. ## Core Features & Use Cases - Reproduction-First Workflow: Measure the actual flake rate with 200+ iteration loops (pytest-repeat, go test -count) before and after any fix, so fixes are proven rather than assumed. - Root-Cause Taxonomy: Systematically classify flakes into interference (polluting tests), time, async ordering, unordered collections, randomness, or CI environment deltas — with concrete fixes for each. - Polluter Bisection: For suite-only failures, replay the CI test order and bisect preceding tests to find the test leaking shared state, then fix the polluter rather than patching the victim. - Use Case: A test fails roughly 1 in 30 CI runs but never locally. You enable local parallelism matching CI, reproduce the flake, trace it to a check-then-insert race on a shared sequence, and ship a unique constraint plus upsert — fixing a production bug the flaky test had been detecting all along. ## Quick Start Diagnose why my test passes locally but fails intermittently in CI and identify the root cause instead of adding a retry.
