Skill Creator

Official Anthropic framework providing structured guidance for developing modular skill packages that extend Claude's capabilities with specialized knowledge, workflows, and tool integrations.

Core Purpose

The Skill Creator framework helps you develop skills that deliver:

• Specialized workflows - Multi-step procedures for specific domains
• Tool integrations - Instructions for file formats or APIs
• Domain expertise - Company-specific knowledge and business logic
• Bundled resources - Scripts, references, and assets for complex tasks

Skill Components

Progressive Disclosure Design

Six-Step Creation Process

1. Understand Usage

Gather Concrete Examples:

• How will this skill be used?
• What specific tasks will it handle?
• Who is the target user?
• What problems does it solve?

Good Starting Point:

• 3-5 real-world usage examples
• Actual prompts users would give
• Expected outputs and behaviors

2. Plan Contents

Identify What You Need:

Scripts:

• Repetitive, deterministic tasks
• File processing operations
• Data transformations
• Validation checks

References:

• Detailed technical documentation
• API specifications
• Best practices guides
• Example code

Assets:

• Template files
• Boilerplate code
• Configuration examples
• Sample outputs

3. Initialize

Use init_skill.py:

python init_skill.py my-skill-name

This Creates:

• Skill directory structure
• SKILL.md template
• README.md
• Optional directories (scripts/, references/, assets/)

4. Edit

Write SKILL.md:

Metadata Section:

---
name: my-skill-name
description: Clear, concise description of what this skill does
when_to_use: Specific triggers that should activate this skill
---

Instructions Section:

• Use imperative/infinitive form ("Create...", "Analyze...", "Generate...")
• Be objective and instructional
• No second-person directives ("you should...")
• Focus on what to do, not why

Good Example:

## Image Processing

Resize images while maintaining aspect ratio:
1. Read image dimensions
2. Calculate new dimensions
3. Apply scaling transformation
4. Save resized image

Bad Example:

## Image Processing

You should resize images carefully to make sure you maintain
the aspect ratio so that they don't look distorted.

5. Package

Use package_skill.py:

python package_skill.py my-skill-name

This:

• Validates SKILL.md structure
• Checks metadata completeness
• Verifies file organization
• Creates distribution package (.zip)

6. Iterate

Test and Refine:

• Use skill in real scenarios
• Gather user feedback
• Identify missing information
• Update documentation
• Re-package and distribute

Writing Standards

name: financial-modeling
description: Create Excel-based financial models with formulas
when_to_use: >
  When user requests financial projections, DCF models,
  budget templates, or three-statement models

Instruction Language

Use Imperative/Infinitive Form:

• ✅ "Calculate revenue projections"
• ✅ "Generate test cases"
• ✅ "Validate input data"
• ❌ "You should calculate revenue projections"
• ❌ "We need to generate test cases"

Be Objective and Instructional:

• ✅ "Load configuration from config.json"
• ✅ "Apply validation rules to user input"
• ❌ "Remember to load the config file"
• ❌ "Don't forget to validate input"

Reference Material Placement

Best Practices

Skill Scope

Good Skill Scope:

• Focused on specific domain or task
• Clear activation triggers
• Self-contained functionality
• Reasonable complexity

Too Broad:

• "general programming help"
• "solve any business problem"
• "answer questions"

Too Narrow:

• "add two numbers"
• "capitalize text"
• Built-in Claude capabilities

Documentation Quality

Effective Documentation:

• Clear, actionable instructions
• Concrete examples
• Step-by-step procedures
• Expected inputs/outputs documented

Poor Documentation:

• Vague descriptions
• Missing examples
• Unclear activation conditions
• Incomplete procedures

Resource Organization

Well-Organized:

my-skill/
├── SKILL.md (main instructions)
├── README.md (user-facing docs)
├── scripts/
│   ├── validate.py
│   └── process.py
├── references/
│   ├── api-docs.md
│   └── examples.md
└── assets/
    └── template.json

Common Patterns

Script-Heavy Skills

When to Use:

• Repetitive file operations
• Data transformations
• Validation tasks
• Deterministic workflows

Structure:

• Minimal SKILL.md (workflow overview)
• Comprehensive scripts/
• Usage examples in references/

Knowledge-Heavy Skills

When to Use:

• Domain expertise encoding
• Best practices guides
• Specialized methodologies
• Complex decision frameworks

Structure:

• Detailed SKILL.md (core knowledge)
• Extensive references/ (deep content)
• Few or no scripts

Tool Integration Skills

When to Use:

• Working with specific file formats
• API integrations
• Third-party tool workflows
• Specialized software interaction

Structure:

• SKILL.md (integration patterns)
• scripts/ (helper utilities)
• references/ (API/tool docs)
• assets/ (examples, templates)

Validation Checklist

Before packaging:

• Metadata clearly states when to activate
• Description is concise and accurate
• Instructions use imperative language
• SKILL.md is under 5,000 words
• Detailed content moved to references/
• Scripts are documented with usage examples
• Assets have clear purposes
• README provides user-facing documentation
• Tested with real usage scenarios

Distribution

After Packaging:

1. Share .zip file
2. Provide usage instructions
3. Document known limitations
4. Specify dependencies
5. Include version number
6. Maintain changelog

About This Skill

Official Anthropic framework for developing modular skill packages that extend Claude's capabilities with specialized knowledge, workflows, and tool integrations.