| name | verify-test-quality |
| description | Apply five concrete heuristics to detect tautological tests, mock-rubber-stamps, and tests written after production code. Use after writing a test, before running it, and any time a test passes too easily. |
Verify test quality
Apply these five heuristics to the test you just wrote (or are reviewing). Any that triggers warrants a closer look.
Heuristics
-
Test imports SUT but only asserts on mock return values.
If the test mocks the SUT's collaborators and then asserts that the SUT returns what the mocks returned, the test isn't testing the SUT — it's testing the mock framework. Restructure so assertions land on values the SUT transforms, not values it passes through.
-
Test body has no expect calls on SUT-returned data.
Tests that only assert on mocks (expect(mockFn).toHaveBeenCalledWith(...)) without ever inspecting what the SUT returned are testing the call shape, not the behavior. Add an assertion on the SUT's return value.
-
Test name matches /should call/ or /should return what mock returns/.
These are red flags. Rename to describe the observable outcome. (See: derive-test-name-from-behavior.)
-
Test was created AFTER the production code it claims to test.
Use file_edits ordering to check. If the test file was edited after the production file in the same TDD cycle, the test was written to match the implementation — the test cannot fail in a way that catches the implementation being wrong. The killer signal.
-
Coverage delta of new test is zero.
A test that doesn't increase coverage either duplicates an existing test or asserts something the type system already enforces. Re-evaluate.
Reusable outside TDD
Code review and coverage-gap workflows use this. The PostToolUse post-tool-use/test-quality.sh hook (orchestrator-scoped) is a narrow automated check today: it scans test-file edits for escape-hatch tokens (it.skip/it.todo/it.fails/it.concurrent, the test.* and describe.* equivalents, and .skipIf/.todoIf) and records tdd_artifacts(kind='test_weakened') when any of them appear. Heuristics #1, #2, and #4 — mock-rubber-stamping, no-SUT-assertion, and test-after-code — are NOT detected by the hook; they remain agent/reviewer responsibilities and are why this skill is consulted manually after writing a test. Heuristic #3 (suspicious test names) is a manual review item too. Heuristic #5 (zero-coverage-delta) requires a coverage run to evaluate.
