| name | evaluate |
| description | Run quality evaluation on the current solution. Grades every completed feature against acceptance criteria on Completeness, Correctness, Code Quality, Test Coverage, and Architecture Fit. Marks failures as needs-rework. |
Evaluate Solution
Run the full evaluator protocol:
- Read
docs/PLAN.md, docs/FEATURES.json, docs/PROGRESS.md, and docs/ARCHITECTURE.md
- For every feature marked
"complete":
- Verify each acceptance criterion against the actual code
- Run tests if they exist
- Score on Completeness, Correctness, Code Quality, Test Coverage, Architecture Fit (1-5 each)
- Hard threshold: any score ≤ 2 means FAIL
- Write the evaluation report to
docs/EVALUATION.md
- Update any failed features to status
"needs-rework" in docs/FEATURES.json
- Summarize results and recommend next steps
Be skeptical. The builder tends to approve its own work — your job is to catch what it missed.
Grading Criteria
| Criterion | 1 (Fail) | 3 (Adequate) | 5 (Excellent) |
|---|
| Completeness | Missing core functionality | All criteria met | Sensible additions |
| Correctness | Broken or buggy | Works for happy path | Handles edge cases |
| Code Quality | Messy, inconsistent | Clean and readable | Well-structured |
| Test Coverage | No tests or broken | Core paths tested | Edge cases tested |
| Architecture Fit | Violates boundaries | Follows architecture | Strengthens it |
Calibration Rules
- Don't be lenient — better to flag a false positive than miss a real bug
- Be specific — include file/line references
- Grade against the SPEC, not your preferences