Skill Builder

Skill Builder

This skill helps you create well-structured Claude Code Skills. Use it when the user wants to build a new skill for Claude Code.

When to Use This Skill

Invoke this skill when users:

• Want to create a new Claude Code Skill
• Ask how to build or structure a skill
• Need help with skill frontmatter or organization
• Want to validate an existing skill structure

Skill Creation Workflow

Step 1: Understand the Requirements

Ask the user these questions (use AskUserQuestion tool when appropriate):

1. What will the skill do? - Get a clear description of the skill's purpose
2. When should it trigger? - Understand what user requests should activate it
3. What's the scope? - Is it project-specific or personal/general-purpose?
4. What resources are needed? - Templates, scripts, reference data, etc.

Step 2: Determine Skill Location

Based on scope, choose the directory:

• Project-specific: .claude/skills/[skill-name]/ (within project root)
• Personal/general: ~/.claude/skills/[skill-name]/ (available everywhere)

Step 3: Create Directory Structure

Use this pattern:

skill-name/
├── SKILL.md           (required: main instructions)
├── REFERENCE.md       (optional: detailed reference info)
├── FORMS.md           (optional: templates and forms)
├── scripts/           (optional: executable code)
│   ├── script1.py
│   └── script2.sh
└── resources/         (optional: templates, data files)
    └── template.txt

Step 4: Write SKILL.md

The SKILL.md must follow this structure:

---
name: skill-name
description: Clear description of what this skill does and when to use it
---

# Skill Name

Main instructions for Claude go here.

## When to Use

Describe scenarios where this skill should be invoked.

## How It Works

Step-by-step instructions for Claude to follow.

## Examples

Provide examples of usage.

Frontmatter Requirements:

• name: Max 64 chars, lowercase letters/numbers/hyphens only (e.g., "excel-automation")
• description: Non-empty, max 1024 chars, should clearly indicate when to use the skill
  • IMPORTANT: Include BOTH natural language AND technical keywords
  • Natural language: How non-technical users would describe it ("fix broken code", "save data", "put online")
  • Technical terms: Jargon that developers use ("debug", "database schema", "deploy")
  • Example: "Auto-invoked when user wants to fix broken code, not working, or crashed. Also triggers on debug, stack trace, error analysis."

Best Practices for Instructions:

• Write like an "onboarding guide for a new team member"
• Use clear, step-by-step procedures
• Include contextual examples
• Reference supporting files instead of embedding everything
• Keep instructions under 5k tokens for efficient loading
• Use progressive disclosure: reference detailed info in REFERENCE.md

Step 5: Add Supporting Files (Optional)

REFERENCE.md: Detailed technical information that Claude can read when needed

• API documentation
• Detailed specifications
• Comprehensive examples
• Technical constraints

FORMS.md: Templates and structured formats

• Form templates
• Output format examples
• Structured data patterns

scripts/: Executable code for complex operations

• Python scripts for data processing
• Shell scripts for automation
• Utilities that Claude can invoke via bash

resources/: Static files and templates

• Document templates
• Sample data
• Reference materials

Step 6: Validate the Skill

Check these requirements:

• SKILL.md exists with valid YAML frontmatter
• name is lowercase, alphanumeric with hyphens, max 64 chars
• description is clear, non-empty, max 1024 chars
• Description includes natural language keywords (not just technical jargon)
• Instructions are clear and actionable
• File references are correct
• No internet/API dependencies (Claude Code constraint)
• Only uses pre-installed packages

Run automated validation:

python .validation/validate-skills.py

Validation Scoring (29 points total):

• SKILL.md exists: 1 point
• Valid frontmatter: 3 points
• Name validation: 2 points
• Description validation: 3 points
• Content structure: 5 points
• Token count <5000: 2 points
• Supporting files: 3 points
• Scripts directory: 2 points
• Industry standards: 3 points
• Examples: 2 points
• Natural language keywords: 3 points (NEW)

Passing threshold: 80% (24/29 points)

Step 7: Test the Skill

After creation:

1. Verify the skill is discoverable (Claude should see it in metadata)
2. Test with a sample task that should trigger it
3. Ensure resources load correctly when referenced
4. Check that instructions are clear and complete

Progressive Disclosure Strategy

Design skills to minimize token usage:

1. Always Loaded (Metadata): ~100 tokens

   • Just the YAML frontmatter

2. Loaded When Triggered (Instructions): <5k tokens

   • Core procedures from SKILL.md body

3. Loaded As Needed (Resources): No token cost

   • Files Claude reads via bash commands
   • Scripts Claude executes when required
   • Templates Claude accesses on demand

Common Skill Patterns

Document Generation Skill

doc-generator/
├── SKILL.md          (generation instructions)
├── templates/        (document templates)
└── scripts/          (formatting scripts)

Code Analysis Skill

code-analyzer/
├── SKILL.md          (analysis procedures)
├── REFERENCE.md      (pattern definitions)
└── scripts/          (analysis tools)

Automation Skill

automation-helper/
├── SKILL.md          (automation workflows)
├── FORMS.md          (configuration templates)
└── scripts/          (automation scripts)

Implementation Steps

When helping users build a skill:

1. Gather Information: Use AskUserQuestion to understand requirements
2. Plan Structure: Determine files needed based on complexity
3. Create Directory: Make the skill directory in appropriate location
4. Write SKILL.md: Create with proper frontmatter and instructions
5. Add Supporting Files: Create REFERENCE.md, FORMS.md, scripts as needed
6. Validate: Check all requirements are met
7. Document: Explain to user how to test and use the skill
8. Show Location: Provide the full path to the created skill

Validation Checklist

Before completing, verify:

• Directory created in correct location
• SKILL.md has valid frontmatter
• Name follows naming rules
• Description is clear and helpful
• Instructions are actionable
• Examples are provided
• Supporting files are organized logically
• No external dependencies
• File paths in instructions are correct

Tips for Great Skills

1. Clear Descriptions: Make it obvious when Claude should use the skill
2. Step-by-Step: Write procedures as numbered steps
3. Examples First: Show examples before explaining
4. Minimize Context: Use progressive disclosure extensively
5. Self-Contained: Don't assume external resources are available
6. Tested Procedures: Ensure instructions actually work
7. Helpful Metadata: Write descriptions that help with discovery

Common Mistakes to Avoid

• Using uppercase or special characters in skill name
• Making descriptions too vague or too long
• Embedding all content in SKILL.md instead of using supporting files
• Referencing non-existent files
• Assuming internet access or external APIs
• Creating overly complex skills that should be split
• Forgetting to include "when to use" guidance

Reference Files

See the templates directory for:

• Basic skill template
• Advanced skill template with supporting files
• Example skills for common use cases