claude-skill-creator
Create and structure Claude Code Skills with best practices. Use when building custom Skills, setting up Skill directories, writing SKILL.md files, or creating reusable Claude capabilities.
When & Why to Use This Skill
The Claude Skill Creator is a specialized development framework designed to streamline the creation, structuring, and optimization of custom Claude Code Skills. It provides a standardized approach to building reusable agent capabilities, offering deep guidance on YAML frontmatter configuration, progressive content loading, and directory organization to ensure high-performance AI interactions.
Use Cases
- Standardizing Skill Development: Establish a consistent directory structure and documentation format (SKILL.md, REFERENCE.md) for team-wide or open-source Claude Skill projects.
- Optimizing Token Efficiency: Implement progressive content loading strategies to manage context windows effectively, ensuring Claude only loads detailed instructions when specific triggers are met.
- Improving Agent Reliability: Utilize best practices for writing action-oriented descriptions and defensive examples to reduce hallucinations and ensure the agent triggers the correct skill at the right time.
- Security and Compliance Auditing: Follow built-in security guidelines to audit third-party skills for malicious patterns and ensure safe execution within code environments.
| name | claude-skill-creator |
|---|---|
| description | Create and structure Claude Code Skills with best practices. Use when building custom Skills, setting up Skill directories, writing SKILL.md files, or creating reusable Claude capabilities. |
Claude Skill Creator
A comprehensive guide to creating professional, reusable Claude Code Skills that extend Claude's capabilities with domain-specific expertise.
Quick Start
A basic Skill requires just one file: SKILL.md with YAML frontmatter:
---
name: your-skill-name
description: Brief description of what this Skill does and when to use it
---
# Your Skill Name
## Instructions
Step-by-step guidance for Claude to follow
## Examples
Concrete examples demonstrating the Skill
Skill Structure
Every production-ready Skill should follow this directory structure:
your-skill/
├── SKILL.md # Required: Main instructions with YAML frontmatter
├── README.md # Optional: Human-readable documentation
├── REFERENCE.md # Optional: Detailed API reference or technical docs
├── EXAMPLES.md # Optional: Comprehensive examples and use cases
├── BEST_PRACTICES.md # Optional: Domain-specific best practices
└── scripts/
├── utility.py # Optional: Helper scripts Claude can execute
└── validate.sh # Optional: Validation or setup scripts
Required YAML Frontmatter
Every SKILL.md must include YAML frontmatter with:
name (required)
- Maximum 64 characters
- Only lowercase letters, numbers, and hyphens
- No XML tags
- Cannot contain: "anthropic" or "claude"
- Example:
my-data-processor✓,MyDataProcessor✗
description (required)
- Maximum 1024 characters
- Must be non-empty
- No XML tags
- Should explain BOTH:
- What the Skill does
- When Claude should use it
- Example: "Extract and analyze data from CSV files. Use when the user mentions CSV, data analysis, or spreadsheet processing."
Writing Effective SKILL.md
Structure Your Instructions
1. Overview Section Briefly explain what this Skill enables Claude to do:
## Overview
This Skill enables Claude to create professional presentations from content,
handling layout, formatting, and slide organization automatically.
2. Key Concepts Define essential terms or workflows:
## Key Concepts
- **Slide deck**: A presentation with ordered slides
- **Theme**: Visual styling applied across all slides
- **Layout templates**: Pre-designed slide structures
3. Step-by-Step Workflows Provide clear procedures Claude can follow:
## Creating a Basic Presentation
1. Gather content from user
2. Organize into logical sections
3. Design slide hierarchy
4. Apply consistent formatting
5. Validate structure
6. Generate output
4. Examples Provide concrete, runnable examples:
## Example: Data Report Presentation
User request: "Create a Q4 sales report presentation"
Steps:
1. Create presentation with title slide
2. Add section: "Executive Summary" with key metrics
3. Add section: "Regional Performance" with charts
4. Add conclusion with call-to-action
5. Tools and Techniques Document available tools and patterns:
## Using Python with Your Skill
You have access to:
- pdfplumber: Extract PDF content
- openpyxl: Work with Excel files
- python-docx: Create Word documents
6. Error Handling Explain how to handle common issues:
## Common Issues and Solutions
**Issue**: File format not recognized
**Solution**: Check file extension and try alternative parsers
**Issue**: Missing required data
**Solution**: Request missing fields from user before proceeding
7. Advanced Techniques For complex Skills, include advanced patterns:
## Advanced: Custom Validators
Create validation functions to ensure output quality:
- Syntax validation
- Content completeness checks
- Format consistency verification
Progressive Content Loading
Design your Skill with three levels of content:
Level 1: Metadata (Always Loaded)
---
name: document-processor
description: Process and transform documents. Use for PDF editing, content extraction, format conversion.
---
Token cost: ~100 tokens per Skill
Level 2: Instructions (When Triggered)
Main SKILL.md body with workflows and guidance. Token cost: Under 5k tokens
Level 3: Resources (As Needed)
Reference files, scripts, and additional documentation only loaded when needed. Token cost: Effectively unlimited (no context penalty)
Best Practices
1. Clear, Action-Oriented Descriptions
✓ Good: "Extract tables from PDF documents and convert to CSV format"
✗ Bad: "Does stuff with PDFs"
✓ Good: "Analyze code for security vulnerabilities and generate remediation steps"
✗ Bad: "Code analysis tool"
2. Include When-to-Use Guidance
---
description: Analyze financial data and generate reports with visualizations.
Use when processing spreadsheets, financial data, budget analysis,
or when the user requests charts and trend analysis.
---
3. Provide Defensive Examples
Show what NOT to do:
## Example: Incorrect vs Correct Usage
**Don't**:
- Process files without checking existence
- Skip validation of external input
**Do**:
- Always verify file paths exist
- Validate and sanitize user input
- Provide clear error messages
4. Document Dependencies
## Prerequisites
This Skill requires:
- Python 3.9+
- Libraries: openpyxl, pandas, matplotlib
- Access to filesystem for reading/writing files
5. Create Focused Skills
Each Skill should have one clear purpose:
✓ Good: "email-template-generator" - Creates reusable email templates
✗ Bad: "do-everything" - General-purpose task handler
6. Design for Reusability
Write instructions that work across multiple scenarios:
## Template Generation
The workflow adapts to different template types:
- Transactional emails (confirmations, receipts)
- Marketing emails (newsletters, promotions)
- Notification emails (alerts, reminders)
- Administrative emails (reports, updates)
Security Considerations
What a Malicious Skill Can Do
- Invoke tools in unintended ways
- Execute arbitrary code
- Access sensitive data
- Exfiltrate information
Protection Strategies
- Only use Skills from trusted sources
- Audit all bundled files before using
- Check for suspicious patterns:
- Unexpected network calls
- Unusual file access patterns
- Operations not matching Skill's stated purpose
- Be cautious with external URLs - fetched content may be compromised
Runtime Constraints
Skills run in a code execution container with these limitations:
| Constraint | Details |
|---|---|
| No network access | Cannot make external API calls |
| No package installation | Only pre-installed packages available |
| Filesystem access | Can read/write within container |
| Available packages | See code execution documentation |
| Bash commands | Full bash access for scripting |
Available Tools and Packages
Pre-installed Packages
- Python scientific stack: pandas, numpy, scipy, scikit-learn
- Document processing: openpyxl, python-docx, pdfplumber, reportlab
- Data tools: matplotlib, seaborn, plotly
- Utilities: requests, beautifulsoup4, lxml, PyYAML
File Operations
- Read/write files with Python or bash
- Execute shell scripts
- Create temporary files for processing
Code Execution
- Python scripts (any version)
- Bash commands and scripts
- Chained operations
Sharing Your Skill
On Claude.ai
Upload as zip file through Settings > Features (Pro/Max/Team/Enterprise with code execution)
Via Claude API
Upload using /v1/skills endpoints - available workspace-wide
In Claude Code
Filesystem-based in .claude/skills/ - personal or project-specific
GitHub Public Repository
Package as open-source with comprehensive README and examples
Testing Your Skill
Before publishing, verify:
Metadata validation
- Name follows format rules
- Description is clear and includes when-to-use guidance
- No XML tags in YAML
Instructions clarity
- Step-by-step workflows are clear
- Examples are concrete and runnable
- Error cases are documented
Functionality
- Test main workflow
- Test error conditions
- Verify output quality
Documentation
- README is comprehensive
- Examples work as written
- All dependencies documented
Common Skill Patterns
Pattern 1: Data Transformation
Transform input data from one format to another:
Input → Validate → Transform → Output
Pattern 2: Content Analysis
Analyze content and generate insights:
Input → Parse → Analyze → Summarize → Report
Pattern 3: Document Generation
Create new documents from templates and data:
Template + Data → Combine → Format → Output
Pattern 4: Workflow Automation
Execute multi-step processes:
Trigger → Step 1 → Step 2 → ... → Step N → Complete
Next Steps
- Define your Skill's purpose - What specific capability does it add?
- Create SKILL.md with clear instructions and examples
- Organize supporting files in a logical structure
- Test thoroughly with various inputs
- Document everything for other users
- Share responsibly - secure upload and clear licensing