Spec-driven workflow covering requirement gathering, spec authoring, implementation review, and verification — with skills, rules, and evaluation scenarios.
96
90%
Does it follow best practices?
Impact
98%
1.19xAverage score across 9 eval scenarios
Passed
No known issues
Review completed implementation to confirm all spec requirements are satisfied and specs remain accurate.
Find relevant specs. Identify all specs whose targets: match the files changed during implementation:
# List changed files, then find specs targeting them
git diff --name-only HEAD~1 | head -20
grep -rl "targets:" specs/*.spec.mdCheck each requirement. Walk through every requirement in the spec. For each one:
src/auth.py:42)[@test] link, confirm the test exists and exercises the requirementRun linked tests. Execute all [@test] paths and record results:
grep -rh '\[@test\]' specs/*.spec.md | sed 's/.*\[@test\] *//' | sed 's/`$//'If any test fails, the review cannot pass — diagnose before continuing.
Capture ALL discovered requirements. Compare the implementation against the spec and identify every behavioral difference — not just new features, but also:
For each discovered requirement:
Update spec metadata. If implementation changed the file structure:
targets: for new, moved, or deleted files[@test] links for restructured testsProduce review summary:
## Review: [Spec Name]
### Requirements
- [x] Requirement 1 — implemented in src/foo.py:42
- [x] Requirement 2 — implemented in src/bar.py:18
- [ ] Requirement 3 — MISSING: no error handling for empty input
### Discovered requirements
- Added: timeout handling for API calls (not in original spec)
### Test results
- Linked tests: 5 passed, 0 failed
- New tests added: tests/test_timeout.py
### Spec updates
- Added timeout requirement to section 3
- Updated targets to include src/baz.pyThe review passes when all requirements are checked, all linked tests pass, and no discovered requirements are left undocumented.