# Agent Skills - Standard Protocol for AI Agents

## Overview
Agent Skills is a standardized protocol for giving AI agents new capabilities, domain expertise, and repeatable workflows. It aims to bridge the gap between static knowledge and active agent behavior by providing a universal format for defining skills.

## Core Concepts

### Domain Expertise
Package specialized knowledge into reusable instructions. Examples include:
- Legal review processes
- Data analysis pipelines
- Compliance checks

### New Capabilities
Give agents explicit capabilities they wouldn't otherwise have:
- Creating presentations (PPTX, PDF)
- Building and deploying MCP servers
- Analyzing complex datasets

### Repeatable Workflows
Turn multi-step tasks into consistent, auditable workflows. This ensures that agents follow specific procedures reliable every time.

### Interoperability
The core promise of Agent Skills is reuse. Write a skill once and use it across any skill-compatible agent product or ecosystem.

## Adoption & Ecosystem
The standard is supported by or compatible with:
- Cursor
- Amp
- Letta
- Goose
- GitHub
- VS Code
- Claude Code
- Claude
- OpenAI Codex

## Technical Specification
Skills are defined in a structured format (typically YAML/JSON or Markdown with frontmatter) that defines:
- Inputs: What the skill needs to run.
- Steps: The sequence of actions or prompts.
- Outputs: What the skill returns.

## Development
Agent Skills is an open development project. 
- GitHub: https://github.com/agentskills/agentskills
- Website: https://agentskills.io