How it Works
The Agent Skills protocol defines a simple lifecycle for how agents interact with skills:
1. Discovery
At startup, agents load only the lightweight metadata (name and description) of each available skill. This keeps the context window clean and efficient.
2. Activation
When a user's request matches a skill's description, the agent "activates" it by reading the full instruction set from the skill's definition file.
3. Execution
The agent follows the step-by-step instructions provided by the skill, which may involve running code, processing files, or calling external APIs.
Directory Structure
Skills are organized in a simple directory structure. Each skill lives in its own folder and requires a single `SKILL.md` file correctly.
skill-name/
└── SKILL.md # Required: The main definition file
└── scripts/ # Optional: Helper scripts (Python, Node, etc.)
└── references/ # Optional: Text files for RAG
└── assets/ # Optional: Images or templatesThe SKILL.md Standard
At the heart of every skill is a `SKILL.md` file. It combines YAML frontmatter for metadata with human-readable Markdown for instructions. This "Dual-Audience" format serves both the AI agent and human developers.
---
name: pdf-processing
description: Extract text and tables from PDF files.
---
# PDF Processing
## When to use this skill
Use this skill when the user needs to work with PDF files...
## How to extract text
1. Use `pdfplumber` for text extraction...Why use Agent Skills?
Self-documenting
The instructions for the agent ARE the documentation for the developer.
Portable
Skills are just files. They are easy to version control, share, and edit.
Universal
Works with any LLM and any agent framework that supports the specification.
Secure
Sandboxed execution and explicit permission scopes.
Start Exploring
Ready to see what's possible? Browse the marketplace to discover hundreds of community-contributed skills.