SkillsLLM
CategoriesBlogAI NewsAbout
HomeBlogHow to Create a SKILL.md File

How to Create a SKILL.md File

6/15/2025by SkillsLLM Team

A SKILL.md file is the heart of any AI skill. It tells AI assistants what your tool does, how to use it, and when to apply it. Here's how to write one that works.

Basic Structure

Every SKILL.md should include these sections:

# Skill Name

Brief description of what this skill does.

## When to Use

Describe the situations where this skill is helpful.

## Instructions

Step-by-step instructions for the AI assistant.

## Examples

Show concrete examples of input and expected output.

Writing Effective Instructions

The key to a great SKILL.md is clarity. AI assistants follow instructions literally, so be specific:

Bad: "Help the user with testing"

Good: "When the user asks to write tests, use the Jest testing framework. Create test files in the __tests__ directory with the naming convention [component].test.tsx. Always include at least one positive test case and one edge case."

Adding Context

You can include additional context that helps the AI understand your project:

  • File patterns — tell the AI where to find relevant files
  • Conventions — describe naming conventions, code style, and patterns
  • Dependencies — list required packages or tools
  • Constraints — mention any limitations or things to avoid

Tips for Quality

  1. Be specific — vague instructions lead to vague results
  2. Include examples — AI assistants learn best from concrete examples
  3. Test iteratively — try your skill, refine the instructions based on results
  4. Keep it focused — one skill should do one thing well
  5. Document prerequisites — list any setup steps or dependencies

Submitting to SkillsLLM

Once your SKILL.md is ready and pushed to GitHub:

  1. Sign in to SkillsLLM with your GitHub account
  2. Click "Submit Skill" in the header
  3. Paste your repository URL
  4. Your skill will be automatically scanned for security and listed in the marketplace

The security scan checks for common vulnerabilities like command injection, hardcoded secrets, and unsafe dependencies. Fix any issues before resubmitting.