Constitution Compliance Review Skill

Priorities

Clarity > Accuracy > Actionability

Clarity matters most because a confusing score helps no one. Accuracy ensures the score reflects actual alignment. Actionability means every finding should suggest concrete improvements.

Goal

Evaluate plugin files (commands, agents, skills) against the Anthropic Constitution's principle of reasoning-based over rule-based instructions. Produce a structured report with an overall alignment score (1-10), section-by-section breakdown, and specific transformation suggestions.

This skill helps plugin authors understand where their prompts fall on the rule/reasoning spectrum and how to improve them.

What Qualifies as Alignment

The Constitution establishes that Claude performs better with judgment criteria and reasoning than with rigid rules. Alignment means:

High alignment (7-10/10):

• Explains why constraints exist, not just what they are
• Provides judgment criteria for edge cases rather than exhaustive if/then rules
• Describes desired outcomes and checkpoints rather than step-by-step procedures
• Trusts the agent to use its intelligence for format and style choices
• Uses rules sparingly, only where error costs are severe or predictability is critical

Mixed alignment (4-6/10):

• Some reasoning provided, but mixed with rigid procedures
• Rules exist with partial explanations
• Judgment criteria present but undermined by conflicting rigid templates
• Over-specifies formatting or process details

Low alignment (1-3/10):

• Mostly or entirely rule-based instructions
• Step-by-step procedures without reasoning
• Rigid templates without flexibility
• Rules without explanation of their purpose
• Excessive enumeration of cases the agent could handle with judgment

How to Conduct a Review

Accept a file path as input. If no path provided, ask the user which file to review.

1. Read the file using the Read tool
2. Analyze section by section:
   • Identify the purpose of each section
   • Classify each as reasoning-based, rule-based, or hybrid
   • Note specific patterns (explained constraints, rigid procedures, judgment criteria, etc.)
3. Score on the 1-10 scale using the rubric in references/scoring-rubric.md
4. Identify specific transformation opportunities:
   • Quote the rule-heavy sections
   • Explain why they're rule-based
   • Suggest reasoning-based alternatives
5. Output the structured report

Use your judgment about how granular to get. For a small file, review every section. For a large file, focus on the most impactful sections.

When to Apply Strict vs Lenient Scoring

Be strict when:

• The file is a high-frequency command (commit, review, research) that shapes user experience
• Rule-based patterns dominate the file
• The consequences of brittleness are high (safety checks, git operations, destructive commands)

Be lenient when:

• The file includes some rules but provides reasoning for them
• The domain genuinely requires predictability (security, error handling)
• The author is clearly trying to balance judgment and structure

The goal is improvement, not perfection. A score of 7/10 is excellent. A score of 9/10 is exceptional.

Output Format

Structure your review as follows:

Overall Assessment

• Alignment Score: [1-10]/10
• Classification: [Reasoning-based / Mixed / Rule-based]
• Summary: [1-2 sentence assessment]

Section-by-Section Breakdown

For each major section:

• Section: [Section name or quote]
• Score: [1-10]/10
• Pattern: [Reasoning-based / Rule-based / Hybrid]
• Reasoning: [Why this score? What patterns did you observe?]

Transformation Suggestions

For each rule-heavy section identified:

Current (rule-based):

[Quote the actual text]

Why this is rule-based: [Explanation]

Suggested (reasoning-based):

[Rewritten version with reasoning]

Why this is better: [Explanation of the improvement]

Strengths

What is this file doing well? Quote specific examples of reasoning-based instruction.

Final Recommendation

Should this file be rewritten? If so, what sections are highest priority?

References

• references/scoring-rubric.md — 1-10 alignment scale with examples at each level
• Anthropic Constitution (January 2026) — "We generally favor cultivating good values and judgment over strict rules and decision procedures" (p. 5)

Arguments

$ARGUMENTS