evaluator

Reference Files

Detailed evaluation guidance:

• evaluation-criteria.md - Correctness, clarity, and effectiveness standards for all customization types
• evaluation-process.md - Step-by-step validation process from identification to reporting
• report-format.md - Standardized report template and guidelines
• common-issues.md - Frequent problems by type with prevention best practices
• examples.md - Good vs poor customization comparisons with assessments

---

Focus Areas

• YAML Frontmatter Validation - Required fields, syntax correctness, field values
• Markdown Structure - Organization, readability, formatting consistency
• Tool Permissions - Appropriateness of allowed-tools, security implications
• Description Quality - Clarity, completeness, trigger phrase coverage
• File Organization - Naming conventions, directory placement, reference structure
• Progressive Disclosure - Context economy, reference file usage
• Integration Patterns - Compatibility with existing customizations, settings.json health

Approach

When evaluating a Claude Code customization, this skill follows a systematic process:

1. Read and parse the target file(s) to extract structure and content
2. Validate YAML frontmatter for required fields and correct syntax
3. Apply type-specific validation criteria (agent/skill/command/hook/output-style)
4. Assess context economy and progressive disclosure usage
5. Verify tool permissions match actual tool usage
6. Check integration with settings.json and other customizations
7. Generate structured report with specific findings and recommendations
8. Prioritize issues by severity (correctness > clarity > effectiveness)

Detailed criteria, process steps, and examples are available in the reference files above.

Tools Used

This skill uses read-only tools for analysis:

• Read - Examine file contents
• Grep - Search for patterns across files
• Glob - Find files by pattern
• Bash - Execute read-only commands (ls, cat, head, tail, git log, etc.)

No files are modified during evaluation. Reports can be saved to ~/.claude/logs/evaluations/ by the invoking command or skill.