← ClaudeAtlas

test-auditlisted

Use when auditing, pruning or sweeping a test suite, when deciding whether a new or changed test earns its place, when a test passes but would not catch the bug it is named for, or when production exports exist only so tests can reach them.
sneg55/agent-starter · ★ 75 · Testing & QA · score 88
Install: claude install-skill sneg55/agent-starter
# Test Audit A test earns its place by failing on a credible regression that no stronger test already catches. This skill has two uses: the **authoring gate** for every new or changed test, and the **audit** for an existing suite. ## Authoring gate Before adding or changing a test, answer all four: 1. Which observable behavior, invariant or independent contract does it protect? 2. Which credible regression makes it fail? 3. Why does existing proof miss it? Each contract has one primary owner at the strongest boundary. A second layer needs a distinct transport or lifecycle risk. Extend an existing table or fixture before writing a new file. 4. Does it need an export, flag, wrapper or injection hook that production never calls? Then move the proof to the real boundary instead. A regression test must fail before the fix for the intended reason and pass after it. A behavior-preserving refactor must not break it. ## Junk patterns - Assertion-free probes, self-comparisons, expected values computed by the subject. - Copied inventories of exports, routes, error ids or constants that the type checker already enforces. - Private-predicate or call-shape tests that duplicate a boundary test; the same shared helper replayed at every layer it crosses. - Mocks or fakes that implement the behavior being asserted. - Fixtures that fabricate receipts, ordering or persistence the owner should produce. - Negative tests that pass on an unrelated guard. Check what rejects the in