---

name: review-spec description: Validate feature specs for relevance, edge cases, and compatibility with existing patterns. Use after creating a spec. user-invocable: false context: fork agent: general-purpose model: sonnet

Review Spec

Input: $ARGUMENTS

Process

1. Read the spec file

2. Read existing patterns

   • CLAUDE.md (project root and .claude/)
   • .claude/rules/*.md
   • Any referenced "Related files" in the spec

3. Validate against criteria

Relevance

• Spec stays focused on the feature
• No tangential details or scope creep
• Clear boundaries on what's in/out

Edge Cases

• Error states covered
• Boundary conditions addressed
• Failure modes considered

Architecture Quality

• No code smells in proposed architecture
• Abstractions are appropriate (not over/under-engineered)
• Integration points are clean

Library Usage

• If external libraries mentioned, verify idiomatic usage
• Flag outdated patterns or deprecated approaches
• Check compatibility with existing stack

Abstraction Level

• Behavioral/contractual, not implementation details
• No pseudocode or type definitions sneaking in
• High-level architecture is fine, micro-details are not

Pattern Compatibility

• Doesn't contradict CLAUDE.md guidance
• Respects existing rules in .claude/rules/
• Fits with codebase conventions found in related files

Context Coverage

• If spec spans multiple domains (10+ files, multiple layers)
• Check .claude/context/ for corresponding context documents
• Flag if large spec lacks context docs for planning
• Each major domain should have its own context file

Output

If no issues found:

CopyPASS

If issues exist, numbered plaintext feedback:

Copy1. Scope: [what strays from the feature]
2. Edge case: [missing error/boundary handling]
3. Architecture: [code smell or design issue]
...