test-design-review

Point the review at the test files, diff, or pull request you want assessed. It evaluates every rule in the test-design rubric, keeps a ledger of findings, and reports violations with location, risk, and remedy ranked by severity.

Assertions on mechanisms rather than outcomes are brittle: mock-recording checks like have_received, reads of cache keys, and reach-ins via .send or instance_variable_set. Assert the observable end result instead so the test survives implementation changes.

Yes, but only when explicitly asked. By default it stops after publishing the ledger and findings report; if you request fixes, it applies them and re-runs the affected tests.

Record order is not guaranteed, so .first and .last depend on incidental ordering that can silently change. Replace them with explicit queries using where, or assert on count changes, to make the test deterministic.

The review asks you to name a target. It never infers scope from a branch or default ref, so you must supply files, a diff, revisions, or a pull request explicitly.