Test Review

Test Review

Review test adequacy for changed code. For large changes spanning many files or test suites, fan out fast-tier readers per logical area — one per package or suite — each returning gap candidates. Synthesize and filter in this session.

Scope

1. Coverage gaps

• new exported behavior with a plausible failure mode — name the bug a missing test would catch
• new branches or error paths without coverage
• changed behavior that existing tests do not exercise
• renamed concepts or protocol terms without tests that cover the new form and guard the old form if needed

2. Edge cases

• boundary conditions (empty inputs, missing files, null/undefined)
• error recovery paths (command not found, parse failures)
• configuration variants the code handles

3. Test quality

• tests asserting implementation details instead of behavior (method call sequences break on refactor)
• tests duplicating coverage without distinct scenarios
• fragile tests (timing, ordering, absolute paths)
• missing cleanup (temp files, cache state)
• mocking internals instead of testing through the real contract — mock at boundaries only
• test names that don't read as specifications
• tests with more than one reason to fail — assert one behavior each

4. Unnecessary tests

• tests covering trivial pass-through or type-only modules
• tests duplicating what the type system guarantees
• tests for code that has since changed

Evidence threshold

Flag a gap as Must-add or Should-add only if you can name the concrete bug or regression the test would catch. If you can't name the bug, don't flag it. Never demand 100% coverage.

• Low-value (don't flag): formatLabel() has no test — trivial pass-through, no failure mode.
• High-value (flag): parseConfig() on an empty file returns undefined but callers assume an object — untested path crashes startup.

Output

For each finding: label (Must-add | Should-add | Optional | Remove), source file + test file, what is untested, why it matters (the concrete bug the test would catch), fix direction. When invoked from review, map Must-add → Critical/Fix, Should-add → Fix, Optional/Remove → Consider.

Group the summary by those labels. Cap Must-add/Should-add at ~5; fold the rest into one Optional line or omit. If nothing clears the threshold, say "No test findings".

See also

• tdd for implementation-time test discipline
• review for multi-axis merge decisions

Red flags

• Demanding 100% line coverage
• Flagging missing tests for trivial functions
• Suggesting tests that only verify the type system
• Broad test rewrites instead of targeted additions
• Confusing test quantity with quality

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: cniska
• Source: cniska/skills
• License: MIT

Install and usage instructions live in the source repository linked above.