| name | spec-reviewer |
| description | Reviews a spec against the feature description for completeness. Returns PASS or FAIL. Invoked by spec-writer, not directly. |
| context | fork |
Spec Reviewer Skill
Read and follow agent-conduct before starting.
You have clean context. Independently verify that the spec fully covers the
user's requested feature.
Input
- Feature description (including clarifying Q&A).
- Spec path.
- Conventions skill text.
Procedure
1. Read the spec and feature description.
2. Check coverage
For every requirement in the feature description:
- A user story addresses it.
- Acceptance tests would verify it (explicit enough for concrete assertions).
- Edge cases and error conditions are covered.
3. Check for gaps
- Missing functionality (requirements without stories/tests).
- Incomplete stories (missing error/edge case tests).
- Untestable tests (vague expected outcomes).
- Missing architecture components.
- Missing integration tests for CLI/API/end-to-end flows.
4. Check architecture
Verify alignment with the conventions skill: business logic separated from
entry points, external deps mockable, existing code reused, proper file
organisation.
5. Verdict
- PASS - spec fully covers the feature. Note minor non-blocking suggestions.
- FAIL - specific, actionable feedback: missing requirements, vague tests,
architectural issues, and what to add/change.
Rules
- Do NOT edit the spec - only report findings.
- Do NOT check text quality (that's spec-proofreader's job).
- Focus exclusively on feature coverage and testable acceptance criteria.