What problem does it solve? Writing tests that actually catch bugs is hard: suites often assert values mirrored from the implementation, never fail, and give false confidence. This Skill enforces a disciplined design workflow where every test must demonstrably fail under a plausible production bug before it counts. ## Core Features & Use Cases - Target discovery (Stage 0): When no target is named, it inspects git diffs, coverage artifacts, and untested production files, then ranks candidates by risk bands (auth, money, state machines, time, I/O) to pick what to test. - Mutation-verified RED: The red_proof.py script mutates one production line, runs the scoped test command, reverts, and prints both runs — proving the test catches the defect, with a crash-safe journal so a killed run never leaves mutated source behind. - Isolation auditing: scan_uncontrolled_resources.py scans C#/TS test files for uncontrolled time, network, filesystem, environment, randomness, and shared state, downgrading findings to mitigated when fakes (FakeTimeProvider, MSW, fake timers) are present. - Stack extensions: Bundled dotnet (xUnit v3, Shouldly, WebApplicationFactory) and React (bun test, RTL, happy-dom, Playwright) extensions supply runner commands and red-proof command shapes. - Use Case: Ask for tests for a pricing module; the Skill builds a target card, derives oracles independent of the implementation, writes one test at a time, proves each reddens via mutation, and reports a table of behaviors with red evidence. ## Quick Start Ask the agent to design and write tests for a specific file, behavior, or bug — or just say "write some tests" and let it identify the highest-risk untested target.
