Agent Skills: The Open Format for Extending AI Agents

Agent Skills is a lightweight, open format for packaging specialized knowledge and workflows into folders that any compatible AI agent can load on demand. If you build or use AI agents and need them to reliably handle domain-specific tasks, legal reviews, data pipelines, or presentation formatting, this format gives you a portable, version-controlled way to extend their capabilities without vendor lock-in.

What it does

Agent Skills turns procedural knowledge into a reusable, auditable package. Instead of hoping an agent guesses the right steps, you write explicit instructions in a SKILL.md file, bundle any scripts, reference docs, or templates alongside it, and let the agent load only what it needs when a task matches the skill's description. The result is repeatable, consistent behavior across different agent products that support the standard.

Before you start

You need a text editor and a file system. No special tools or runtimes are required to create a skill. The format is folder-based and works with any version control system you already use. To use a skill with an agent, you need a client that supports the Agent Skills format. See the Client Showcase for a list of compatible products.

Skill folder structure

Every skill is a folder with a required SKILL.md file and optional subdirectories for supporting resources. The standard layout looks like this:

my-skill/
├── SKILL.md          # Required: metadata + instructions
├── scripts/          # Optional: executable code
├── references/       # Optional: documentation
├── assets/           # Optional: templates, resources
└── ...               # Any additional files or directories

The SKILL.md file is the only mandatory piece. It contains metadata (at minimum a name and description) and the instructions the agent follows. The other directories are conventions, not requirements. You can add any files or folders that make sense for your use case.