Response Rater Skill
Step 1: Define Rating Rubric
Use appropriate rubric for the content type:
For Plans:
Dimension Weight Description
Completeness 20% All required sections present
Feasibility 20% Plan is realistic and achievable
Risk Mitigation 20% Risks identified with mitigations
Agent Coverage 20% Appropriate agents assigned
Integration 20% Fits with existing systems
For Responses:
Dimension Weight Description
Correctness 25% Technically accurate
Completeness 25% Addresses all requirements
Clarity 25% Easy to understand
Actionability 25% Provides clear next steps
Step 2: Evaluate Each Dimension
Score each dimension 1-10:
Dimension Scores
Completeness: 8/10
- Has objectives, steps, and timeline
- Missing risk assessment section
Feasibility: 7/10
- Most steps are achievable
- Step 3 timeline is aggressive
Risk Mitigation: 5/10
- Only 1 risk identified
- No mitigation strategies
Agent Coverage: 9/10
- All steps have assigned agents
- Good agent-task matching
Integration: 8/10
- Uses existing APIs
- Minor compatibility concerns
Step 3: Calculate Overall Score
Average the weighted scores:
Overall = (8×0.2) + (7×0.2) + (5×0.2) + (9×0.2) + (8×0.2) = 7.4/10
Step 4: Generate Recommendations
Provide actionable improvements:
Recommendations
High Priority
- Add risk assessment section with 3-5 risks
- Include mitigation strategies for each risk
Medium Priority
- Extend Step 3 timeline by 2 days
- Add fallback plan for external API dependency
Low Priority
- Add success metrics for each step
Step 5: Make Pass/Fail Decision
Apply minimum score thresholds:
Task Type Minimum Score
Standard 7/10
Enterprise 8/10
Critical 9/10
Decision
Score: 7.4/10 Threshold: 7/10 (Standard) Result: PASS ✅
Plan approved with recommendations for improvement.
</execution_process>
<best_practices>
-
Consistent Rubrics: Use same dimensions for comparable content
-
Clear Thresholds: Define pass/fail criteria upfront
-
Actionable Feedback: Provide specific improvements
-
Prioritize Issues: Order recommendations by impact
-
Document Rationale: Explain score justifications
</best_practices>
Rate this plan for user authentication implementation
Response:
Plan Rating: User Authentication
Scores
- Completeness: 8/10 - Good structure, missing tests section
- Feasibility: 9/10 - Realistic timeline and approach
- Risk Mitigation: 6/10 - Only 1 risk identified
- Agent Coverage: 9/10 - Good agent assignments
- Integration: 8/10 - Compatible with existing auth
Overall: 8.0/10 ✅ PASS
Top Recommendations
- Add risk section with security and dependency risks
- Include test plan for each authentication flow
- Add rollback procedure for failed deployment
</usage_example>
Iron Laws
-
ALWAYS use the same rubric dimensions when rating comparable content — inconsistent dimensions make scores meaningless and prevent valid comparison across sessions.
-
NEVER issue a pass/fail decision without documenting score justification for each dimension — unjustified scores cannot be reviewed, challenged, or improved.
-
ALWAYS apply defined minimum thresholds (7/10 standard, 8/10 enterprise, 9/10 critical) — ad-hoc thresholds produce inconsistent approval gates that erode trust in the rating system.
-
NEVER provide vague recommendations — every recommendation must reference the specific dimension it addresses and state the concrete change required.
-
ALWAYS prioritize recommendations by impact — high-priority items that would materially improve the score must be clearly distinguished from low-impact suggestions.
Anti-Patterns
Anti-Pattern Why It Fails Correct Approach
Using different rubric dimensions for comparable content Scores cannot be compared across sessions; the rating loses its evaluative value Always use the same rubric (plans rubric for plans, responses rubric for responses)
Omitting score justification for individual dimensions Scores without justification cannot be reviewed, verified, or acted upon Document specific evidence for each dimension score (what was present, what was missing)
Setting thresholds arbitrarily per session Inconsistent thresholds invalidate the pass/fail gate; teams lose confidence in approvals Always apply the defined thresholds: 7/10 standard, 8/10 enterprise, 9/10 critical
Providing vague recommendations ("improve quality", "add more detail") Vague feedback cannot be acted upon; no change results from the review Reference the specific dimension, score gap, and required concrete change for each recommendation
Listing recommendations without priority ordering Equal-weight feedback causes raters to address low-impact items first Always order by impact: High (affects pass/fail threshold) before Medium before Low
Memory Protocol (MANDATORY)
Before starting:
cat .claude/context/memory/learnings.md
After completing:
-
New pattern -> .claude/context/memory/learnings.md
-
Issue found -> .claude/context/memory/issues.md
-
Decision made -> .claude/context/memory/decisions.md
ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.