review-tests

name	review-tests
description	Guide for the test_reviewer agent when reviewing test specs in step 6. Auto-loaded when a step-6 task is in progress.
user-invocable	false

Reviewing test specs (Step 6)

You are the last gate before implementation begins. Your job is to ensure test specs are complete and testable — not to judge the implementation.

Review checklist

For each DoD criterion:

• Is there at least one test spec that covers it?
• Does the spec's expected_result directly verify the criterion?

For each test spec:

• Is description specific enough to understand what is being tested without reading the code?
• Is expected_result measurable? (HTTP status + body shape, return value, file content, etc.)
• Could an automated test runner verify this without human judgement?
• Is the happy path covered?
• Is at least one error/edge case covered per criterion?

File context (optional)

You have read-only access to the codebase. If relevant existing code is present, you may check whether specs align with the actual interface — but do not reject specs solely because the code doesn't exist yet (it hasn't been written).

Approval

Approve when every DoD criterion has coverage and all specs are specific and verifiable.

Rejection

Reject when:

• One or more DoD criteria have zero test coverage
• expected_result fields are vague ("no errors", "it works", "200")
• Specs are missing expected_result entirely
• Specs test the same behaviour redundantly without covering other criteria

Your notes on rejection must name the specific criterion or spec that failed the check. Do not give generic feedback.

Example rejection notes:

"DoD criterion 'Unauthenticated requests return 401' has no corresponding test spec. Add a spec for this before approving."