| name | evaluation |
| description | Evaluation patterns — sprint contract format, three-layer verification, scoring rubric references. |
Evaluation Skill
Reference skill for the evaluator agent. Points to authoritative sources for evaluation patterns.
Full Workflow
Read .claude/skills/evaluate/SKILL.md for the complete three-layer verification workflow, execution steps, verdict format, and mode behavior.
References
| File | Contents |
|---|
references/contract-schema.json | Sprint contract JSON schema |
references/scoring-rubric.md | Design scoring rubric (4 criteria, weights, exemplars) |
references/scoring-examples.md | Calibration anchors (score 5, 7, 9) — read before scoring |
references/playwright-patterns.md | Selector patterns and assertion patterns for Layer 2 |
Evaluator Behavioral Rules
These rules are non-negotiable. Deviation invalidates the evaluation.
- Execute every check. Do not skip a check because a related check passed.
- Never rationalize a failure. If the check specifies
status: 200 and you get 201, that is a FAIL.
- Evidence over opinion. Every verdict must cite specific output: response body, screenshot path, line number.
- No partial credit on binary checks. API and Playwright checks are pass/fail.
- Design scores are evidence-based. Cite what you observed, not what you assumed.
- Do not infer intent. If the contract says check X and X is absent, the check fails.
- Run checks in order. Layer 1 before Layer 2 before Layer 3.
- Document every check result, even passing ones.
