What problem does it solve? AI-generated and hastily written test suites often pass while proving nothing: tautological assertions, mock-only tests, hardcoded waits, retries masking races, and coverage theater. This Skill audits test code for 40 named anti-patterns, ranks them by severity, and proves the critical ones by deliberately breaking production code and watching whether the test notices. ## Core Features & Use Cases - Tiered pattern detection: Scans tests for 40 tells across assertions, determinism, test doubles, naming, data, and browser end-to-end code, ranked Tier 1 (test is a lie) through Tier 3 (unreadable). - Mutation-based proof: For every Tier 1 finding, breaks the claimed behavior in production code, runs the single test, and pastes the command and output showing it stayed green. - Framework-specific fixes: Translates each tell into Vitest, Jest, Playwright, and Mocha idioms, including grep patterns suitable for CI. - Use Case: After an AI assistant generates a Playwright suite, run this audit to find that retries: 2 in the config is hiding a real race, prove it with --repeat-each 50 --retries 0, and replace the hardcoded wait with a web-first assertion. ## Quick Start Audit the tests in my current diff against main and report every weak assertion with proof by execution.
