Skill Generator

Purpose

Generate production-ready Claude Code skills from natural language descriptions, existing documentation, or code examples.

When to Use

• User asks to "create a skill" or "generate a skill"
• Converting documentation into skill format
• Scaffolding new skill structure
• Creating skills from examples or patterns

Process

1. Understand Requirements

   • Ask clarifying questions if needed:
     • What should the skill do?
     • What tools/APIs does it need?
     • Who is the target audience?
     • Are there existing docs or examples?

2. Determine Skill Type

   • Generator skill (creates content)
   • Integrator skill (connects to services)
   • Converter skill (transforms formats)
   • Or combination

3. Generate SKILL.md Follow this structure:

   ---
   name: [skill-name]
   description: [What it does + when to use it - max 1024 chars]
   allowed-tools: [optional - list tools]
   ---
   
   # [Skill Name]
   
   ## Purpose
   [2-3 sentences]
   
   ## Process
   1. [Step 1]
   2. [Step 2]
   3. [Step 3]
   
   ## Examples
   [Concrete examples]
   
   ## Best Practices
   [Key guidelines]
   

4. Create Directory Structure

   .claude/skills/[skill-name]/
   ├── SKILL.md              # Main file (3-5KB max)
   ├── reference/            # Optional: Detailed docs
   │   └── [topic].md
   └── scripts/              # Optional: Utility scripts
       └── [script].py
   

5. Apply Best Practices

   • Use gerund form for name: processing-pdfs
   • Write description in third person
   • Keep SKILL.md under 500 lines
   • Use progressive disclosure for large content
   • Include concrete examples

6. Review and Refine

   • Check YAML frontmatter is valid
   • Verify description is specific
   • Ensure clear instructions
   • Test mental walkthrough

Output Format

Always create:

1. SKILL.md - Main skill file
2. Directory structure - Planned reference/ scripts if needed
3. Usage examples - How to invoke the skill
4. Next steps - What user should do next

Example Generation

If user says: "Create a skill for processing Excel files"

Generate:

.claude/skills/
├── excel-processor/
│   ├── SKILL.md
│   ├── reference/
│   │   ├── pandas-api.md
│   │   └── openpyxl-api.md
│   └── scripts/
│       ├── analyze.py
│       └── convert.py

SKILL.md includes:

• Name: excel-processor
• Description: "Processes Excel files including data extraction, format conversion, and analysis. Use when working with .xlsx files, spreadsheets, or tabular data."
• Process for reading, analyzing, converting Excel files
• Examples with pandas and openpyxl
• Links to reference docs for detailed API usage

Quality Checklist

Before presenting to user, verify:

• YAML frontmatter is valid (starts with ---)
• Name uses lowercase, numbers, hyphens only
• Description is specific and includes trigger terms
• Description is under 1024 characters
• Instructions are clear and numbered
• Examples are concrete and practical
• File paths use forward slashes
• Progressive disclosure used if content is large
• No reserved words (anthropic, claude) in name

Common Patterns

For Generator Skills

## Purpose
Generate [output type] from [input type]

## Process
1. Analyze [input]
2. Generate [output] using [best practices]
3. Validate output
4. Present to user

For Integrator Skills

## Purpose
Integrate with [service name]

## Process
1. Understand user's goal
2. Prepare data for [service]
3. Call [service] API/tool
4. Handle response
5. Report results

For Converter Skills

## Purpose
Convert [format A] to [format B]

## Process
1. Read [format A] file
2. Parse structure
3. Transform to [format B]
4. Write output
5. Verify conversion

Progressive Disclosure Guidance

If skill will have >10KB of documentation:

In SKILL.md:

## Quick Reference
[Brief overview and common tasks]

## Detailed Documentation
- **API Reference**: See [reference/api.md](reference/api.md)
- **Examples**: See [reference/examples.md](reference/examples.md)
- **Troubleshooting**: See [reference/troubleshooting.md](reference/troubleshooting.md)

Then create the reference files with detailed content.

Troubleshooting

If requirements are unclear:

• Ask: "What specific task should this skill automate?"
• Ask: "What tools or APIs does it need to work with?"
• Ask: "Who will be using this skill?"

If skill seems too complex:

• Consider splitting into multiple focused skills
• Use progressive disclosure for documentation
• Create subtasks as separate skills

If documentation is large:

• Use progressive disclosure pattern
• Create reference/ directory
• Link to detailed docs from SKILL.md

See Also

• SKILL_REVIEWER.md - Review skills for quality
• SKILL_OPTIMIZER.md - Optimize existing skills
• SKILL_ARCHITECT.md - Complete skill workflow

Sources

Based on:

• CLAUDE_SKILLS_ARCHITECTURE.md
• Official Anthropic skill documentation