review-skills

name	review-skills
description	Review and analyze a skill against best practices for length, intent scope, and trigger patterns
disable-model-invocation	true
metadata	{"internal":true}

Review Skills

Review and analyze a skill against best practices for length, intent scope, and trigger patterns.

Prerequisites

Before analyzing, read these resources to understand skill writing principles:

1. ./references/skill-creator/SKILL.md - Core principles, anatomy, and progressive disclosure
2. references/spec.md - Complete Agent Skills specification (required for compliance checks)
3. references/validate.md - Validation checklist (used in Step 2)
4. ./references/skill-creator/references/workflows.md - Workflow patterns (if relevant)
5. ./references/skill-creator/references/output-patterns.md - Output patterns (if relevant)

Reference Examples from Anthropic (REQUIRED)

You MUST read reference skills from Anthropic's repository before analyzing. This is essential for calibrating your review.

1. Ensure cache is available: Check if ./.cache/anthropics-skills/ exists. If not (or if stale), run:

   python scripts/download_anthropics_skills.py
   

2. Read at least 3 reference skills: Before analyzing, read these SKILL.md files from ./.cache/anthropics-skills/skills/:

   Always read these high-quality examples:

   • pdf/SKILL.md - Well-structured workflow skill with clear triggers
   • docx/SKILL.md - Good example of document processing patterns
   • skill-creator/SKILL.md - Meta-skill showing best practices

   Then read 1-2 skills similar to the one being reviewed:

   • For workflow-based skills: xlsx/SKILL.md, pptx/SKILL.md
   • For tool/API skills: mcp-builder/SKILL.md
   • For creative/design skills: brand-guidelines/SKILL.md, frontend-design/SKILL.md
   • For testing skills: webapp-testing/SKILL.md

3. Note patterns to compare: As you read, note:

   • How descriptions are structured (trigger patterns)
   • Length and depth of SKILL.md body
   • How references are organized and used
   • Balance between brevity and completeness

Steps

Step 1: Receive the Skill to Review

The user must provide a skill folder/path to review. If not provided, prompt:

"Please provide the path to the skill folder you want to review (e.g., .claude/skills/my-skill/)"

Step 2: Validate Skill Structure

Using the validation checklist (references/validate.md), verify the skill passes all basic checks:

1. File Structure: SKILL.md exists
2. Frontmatter Format: Valid YAML between --- delimiters
3. Allowed Properties: Only name, description, license, compatibility, metadata, allowed-tools
4. Name Validation:
   • Hyphen-case only (lowercase, digits, hyphens)
   • No start/end hyphens, no consecutive hyphens (--)
   • Max 64 characters
   • Matches directory name
5. Description Validation:
   • No angle brackets (< or >)
   • Max 1024 characters
   • Non-empty

If validation fails: Stop the review and report the specific validation error(s). The skill must pass basic validation before proceeding with the full review.

Step 3: Read the Skill

Read the complete skill structure:

• SKILL.md (frontmatter and body)
• Any files in references/, scripts/, assets/ directories

IMPORTANT: Only analyze the skill provided by the user.

Step 4: Verify Spec Compliance

Check that the skill follows the Agent Skills specification (references/spec.md). Verify:

Directory Structure

• Skill is in a directory matching the name field
• Contains required SKILL.md file
• Optional directories follow conventions: scripts/, references/, assets/

Frontmatter Compliance

Field	Check
name	1-64 chars, lowercase alphanumeric + hyphens, no start/end hyphens, no --, matches directory name
description	1-1024 chars, non-empty, describes what and when
license	If present, short (license name or file reference)
compatibility	If present, max 500 chars
metadata	If present, string keys to string values
allowed-tools	If present, space-delimited tool list

Body Content

• Markdown format after frontmatter
• Recommended: step-by-step instructions, examples, edge cases
• Under 500 lines (move detailed content to references)

Progressive Disclosure

• Metadata (~100 tokens): name + description loaded at startup
• Instructions (<5000 tokens recommended): SKILL.md body loaded on activation
• Resources (as needed): scripts/references/assets loaded on demand

File References

• Use relative paths from skill root
• Keep references one level deep (avoid deeply nested chains)

If spec violations found: Document them clearly in the review output with specific fixes.

Step 5: Analyze the Skill

Perform analysis in four areas, comparing against the reference skills you read from Anthropic's repository:

A. Length Analysis

Using the progressive disclosure guidelines from skill-creator, evaluate:

• Word count in description field
• Line/word count in SKILL.md body
• Number and size of reference files
• Duplication between SKILL.md and reference files

B. Intent Scope Analysis

Evaluate:

• All intents the skill serves
• Whether skill handles multiple distinct use cases
• Whether splitting would improve triggering accuracy
• Trade-offs: context efficiency vs. maintenance overhead

Questions to answer:

• Does this skill try to do too much?
• Are there distinct user intents that deserve separate skills?

C. Trigger Analysis (CRITICAL)

The description field is the primary triggering mechanism. Evaluate it for three types of triggers:

Trigger Type	What to Check
User INTENT	Does it describe what the user wants to do? (e.g., "deploy", "create", "edit")
TECHNICAL context	Does it mention code patterns, file types, imports? (e.g., "base44.entities.*", ".jsonc files")
Project stack	Does it mention frameworks, tools, file structures? (e.g., "Vite", "Next.js", "base44/")

Check:

• Does description cover both intent-based AND technical triggers?
• Is it specific enough to trigger correctly, but broad enough to not miss cases?
• Are there gaps where the skill might not trigger when it should?
• Does it clearly distinguish from similar skills?

Good trigger pattern example:

ACTIVATE when (1) INTENT - user wants to [action]; (2) TECHNICAL - code contains [patterns], uses [APIs]; (3) CONTEXT - project has [structure/files]

Step 6: Provide Recommendations

Summarize findings with actionable recommendations for:

1. Spec Compliance: What needs to be fixed to follow the spec?
2. Length: What should be trimmed or split?
3. Intent Scope: Should it be split or combined?
4. Triggers: How can the description be improved?

Output Format

## Skill Review: [Skill Name]

### Reference Skills Compared
- [List the 3-5 Anthropic skills you read before this review]

### Summary
[1-2 sentence overview]

### Validation Result
- **Status**: [Pass/Fail]
- **Details**: [Validation output or errors]

### Spec Compliance
- Directory structure: [Pass/Fail - details]
- Frontmatter fields: [Pass/Fail - details]
- Body content: [Pass/Recommendations]
- Progressive disclosure: [Pass/Recommendations]
- File references: [Pass/Recommendations]
- **Assessment**: [Compliant/Partially compliant/Non-compliant]
- **Fixes Required**: [List of specific fixes if any]

### Length Analysis
- Description: X words
- SKILL.md body: X lines / X words
- Reference files: X files
- **Assessment**: [Pass/Needs attention]
- **Recommendations**: [Specific suggestions]

### Intent Scope Analysis
- Intents served: [List]
- **Assessment**: [Focused/Broad/Too broad]
- **Recommendations**: [Split suggestions if applicable]

### Trigger Analysis
- Intent coverage: [Yes/Partial/No]
- Technical coverage: [Yes/Partial/No]
- Stack coverage: [Yes/Partial/No]
- **Assessment**: [Strong/Adequate/Weak]
- **Recommendations**: [Specific description improvements]

### Overall Recommendations
1. [Priority 1 action item - spec compliance fixes if any]
2. [Priority 2 action item]
3. [Priority 3 action item]
4. [Priority 4 action item]