What problem does it solve? When a test in tests/ goes red, wobbles between runs, or stays green without proving anything, it is easy to blame the program first and waste hours. This Skill enforces a cheap-to-expensive diagnostic order — environment, snapshot, contention, deadlines, stand-ins, the test itself, and only then the program — so the true cause is found before any code is changed. ## Core Features & Use Cases - Ordered root-cause elimination: Rules out language/locale mismatches (LANGUAGE=en via run.sh), stale VPM_SCRIPT snapshots, and parallel-run contention before any source reading. - Flakiness and deadline analysis: Uses wobbly.sh reports and the unsteady line to separate contention from real faults, and flags tests waiting on wall-clock deadlines instead of conditions. - False-green detection: Lists the signatures of tests that pass without checking — missing check calls, silent sys.exit(0), timer-chain escape branches, and checks that can never go red. - Use Case: A test is red on the CI builder but green locally. Following the Skill, you discover the builder is ~9x slower and the test died on a 30-second deadline — so you fix the waiting condition instead of raising the timeout. ## Quick Start Ask the AI to take apart the failing test using the test-rot checklist and identify whether the program, the test, or the stand-in is at fault.
