>_

Five Steps to Create a Skill

A step-by-step guide to creating your first Agent Skill.

Step 1: Identify Your Core Need

Before writing any content, clearly define:

  • Problem: What challenge does your skill solve?
  • Audience: Who will use this skill?
  • Scope: What should the skill cover (and not cover)?

Questions to Answer

  1. What tasks will the AI perform with this skill?
  2. What domain knowledge is required?
  3. What behaviors should the AI exhibit?
  4. What common mistakes should be avoided?

Example

Problem: Developers need consistent code review feedback Audience: Development teams using pull requests Scope: Code quality, security, and style - not project management

Step 2: Write the Name Field

Your skill's name should be:

  • Descriptive: Clearly indicate what the skill does
  • Concise: 3-5 words maximum
  • Unique: Distinguish from similar skills

Good Names

  • "React Component Guidelines"
  • "API Security Checker"
  • "Brand Voice Guide"

Bad Names

  • "My Skill" (not descriptive)
  • "The Ultimate Complete Development Guide for Everything" (too long)
  • "Helper" (not specific)

Step 3: Write the Description

The description appears in search results and skill cards. Make it count:

  • First 100 characters: Most important - shown in previews
  • Include keywords: Help users find your skill
  • Be specific: Mention key features or use cases

Template

[Action] for [audience] that [key benefit]. Covers [main features].

Example

Code review guidelines for development teams that ensure consistent, high-quality feedback. Covers security, performance, and style checks.

Step 4: Write the Main Content

Structure your content for clarity and effectiveness:

Context Section

Provide background information:

markdown
## Context

This skill helps AI assistants review code with a focus on:
- Security vulnerabilities
- Performance optimization
- Code style consistency
- Best practice adherence

Guidelines Section

List specific rules:

markdown
## Guidelines

1. Always check for input validation
2. Look for potential SQL injection
3. Verify error handling exists
4. Check for hardcoded credentials
5. Review logging practices

Examples Section

Show expected behavior:

markdown
## Examples

### Security Review Comment
"Consider using parameterized queries here to prevent SQL injection.
The current string concatenation is vulnerable to malicious input."

### Performance Review Comment
"This loop iterates over the entire array for each element.
Consider using a Set or Map for O(1) lookups instead of O(n)."

Step 5: Upload and Test

Create the Repository

  1. Go to GitHub and create a new repository
  2. Add your SKILL.md file to the root
  3. Make the repository public

Submit to Marketplace

  1. Visit /submit on the marketplace
  2. Enter your repository URL
  3. Click "Submit"

Test Your Skill

After submission:

  1. Install the skill in your preferred platform
  2. Test with various prompts
  3. Verify the AI follows your guidelines
  4. Iterate and improve based on results

Testing Checklist

Before publishing, test these scenarios:

Normal Operation

  • [ ] AI correctly follows main guidelines
  • [ ] Examples are applied appropriately
  • [ ] Tone and style match expectations

Edge Cases

  • [ ] Handles unusual inputs gracefully
  • [ ] Doesn't conflict with other skills
  • [ ] Works across supported platforms

Out of Scope

  • [ ] AI correctly defers on unrelated topics
  • [ ] Doesn't overreach skill boundaries
  • [ ] Gracefully handles requests outside scope

Next Steps

Edit this page on GitHub