| name | ds-test-quality-review |
| description | Run a strict review of test quality — whether the code that matters is well tested, and whether the tests are good ones. Reports a findings list by default; `--fix` applies the mechanical, unambiguous fixes (writing or redesigning tests stays reported). |
| disable-model-invocation | true |
When invoked, audit the tests in scope against one governing principle: test what matters, and test it well. Coverage is not the goal — a percentage says nothing about whether a test would catch a real failure, or whether it pins the code in place. Hunt two failures with equal energy: critical code that is under-tested (the bug waiting to ship) and tests that are bad (present and green, but worthless or actively harmful — they lock the design and tax every change).
Report-only by default — the output is a prioritized list, no edits unless --fix is passed. This is the after-the-fact audit; to write tests this way from the start, use /ds-test-mode (alongside normal work) or /ds-tdd-mode (test-first).
Arguments
- Treat positional args as scope (files, directories, globs). With no scope, review the tests covering code changed on the current branch.
--full → review the whole test suite instead of just the tests covering the branch's changes. Explicit positional scope still wins; --full only replaces the no-scope default.- Freeform scope ("the auth module", "the whole suite") is interpreted reasonably.
--fix → after reporting, apply only the findings whose fix is mechanical and unambiguous — deleting a worthless or duplicate test, removing a snapshot that asserts nothing. Writing a missing test or rewriting a design-locking one is judgment work, so it stays report-only. After applying, re-run any build/test/lint check already in the loop and revert any fix that breaks it — or that touched more than the intended mechanical edit. Close with a summary of what was applied and what was left.
What to check
Identify the code worth testing first, then judge the tests against it. Run the suite when it's safe and fast — a test that can't run isn't protecting anything.
1. Is the critical code tested at all? (highest priority.) Find the code that would hurt most if it broke — core domain logic, money/auth/permissions, data integrity, parsing, state machines, anything with real branching — and check it has tests that would actually fail if it regressed. A gap here matters far more than a high coverage number elsewhere. Ignore the trivia: getters, glue, framework behavior, the obvious.
2. Edge cases and failure modes. For the risky logic, are the bugs-that-actually-happen covered — boundaries (empty, zero, one, max, off-by-one), invalid and malicious input, error paths, partial failure, concurrency and ordering? Happy-path-only tests on branching logic are a gap.
3. Test quality — does it test behavior? Good tests exercise real behavior through the public interface and survive a behavior-preserving refactor. Flag tests that:
- mock internal collaborators, or assert on call counts / call order
- reach through a side channel (raw DB query, private method) instead of the interface
- assert nothing meaningful — a real break would still pass them
- recompute the expected value the way the code does (
assert add(a, b) == a + b) — the test can't disagree with the code, so a real break still passes
- are non-deterministic: depend on wall-clock, network, or test order
4. Tests that lock the design (worse than no test). A test coupled to implementation breaks on every refactor that changes nothing observable. It doesn't prevent failures — it taxes change and trains people to ignore red. Mock-the-world, snapshot-everything, and internals-asserting tests belong here. Recommend rewriting against the interface, or deleting.
5. Bloat and noise. Tests of trivial code, duplicated assertions, tests that re-verify the framework or the language itself. They add maintenance and run time without buying confidence.
Not coverage. Do not recommend tests to hit a number. A well-tested core with untested trivia is the correct state, not a finding. If you see coverage-chasing — tests written only to color a line green — call it out as the anti-pattern it is.
Output
A prioritized findings list, in this order:
- Critical/core code with no real test (the failure waiting to ship)
- Missing edge cases and failure modes on risky logic
- Tests that lock the design (implementation-coupled — rewrite or delete)
- Weak tests that wouldn't catch a real break
- Bloat — tests to cut
For each finding:
- Anchor to
file:line.
- State the problem in one line, then the suggested fix — and when the fix is "delete it", say so plainly.
- For a coverage gap, name the specific behavior or edge case that's unprotected, not "needs more tests".
Rules:
- A short high-conviction list beats a long pedantic one. Don't pad it with low-value nits.
- Judge by risk, not count. The question is always "would this catch a real failure, and does it survive a refactor?"
- With
--fix, additionally summarize what was applied vs. left (mechanics in Arguments).
