What are Agent Skills?
Modular capabilities that empower AI agents to interact with the world reliably and securely.
Agent Skills are a simple, open format for giving AI agents new capabilities and expertise. They're like onboarding guides for new hires—packaging procedural knowledge, organizational context, and best practices into modular resources that agents can discover and load dynamically.
Skills are self-contained packages of code (Prompt + Tool) that give an agent a specific ability, like 'Search the Web' or 'Send Email'.
Designed with standard metadata so agents can dynamically 'learn' or 'load' them when a user asks for a related task.
Follows a common file structure (SKILL.md, scripts/) ensuring compatibility across different agent frameworks.
What Can Agent Skills Enable?
Package specialized knowledge into reusable instructions, from legal review processes to data analysis pipelines.
Give agents new capabilities (e.g. creating presentations, building MCP servers, analyzing datasets).
Turn multi-step tasks into consistent and auditable workflows.
Reuse the same skill across different skills-compatible agent products.
Progressive Disclosure: The Core Design Principle
Skills use progressive disclosure, like a well-organized manual that starts with a table of contents, then specific chapters, and finally a detailed appendix. Agents load information only as needed, effectively managing the context window.
Level 1: Metadata
At startup, the agent pre-loads the name and description of every installed skill into its system prompt. This provides just enough information for the agent to know when each skill should be used without loading all of it.
Level 2: SKILL.md
If the agent thinks the skill is relevant to the current task, it will load the skill by reading its full SKILL.md into context. This is the second level of detail.
Level 3+: Reference Files
As skills grow in complexity, they may contain too much context to fit into a single SKILL.md. In these cases, skills can bundle additional files within the skill directory, which the agent can choose to navigate and discover only as needed.
Anatomy of a Skill
Every skill follows a rigorous structure to ensure predictability for the LLM. It's not just code; it's a semantic interface.
- ✓Definition File (SKILL.md)
- ✓Python/Node.js Scripts
- ✓Asset Directory
- ✓Metadata Registry
├── SKILL.md ├── scripts/ │ ├── browser_tool.py │ └── util.py ├── assets/ └── references/
Skill Authoring Best Practices
- ✓Keep it concise - the context window is a public good
- ✓Use progressive disclosure patterns
- ✓Provide clear workflows for complex tasks
- ✓Implement feedback loops and validation steps
- ✓Use consistent terminology
- ✗Avoid deeply nested references
- ✗Don't include time-sensitive information
- ✗Don't offer too many options
- ✗Avoid assuming tools are installed
- ✗Don't use Windows-style paths
Core Principle: Concise is Key
The context window is a public good. Your Skill shares the context window with everything else the agent needs to know, including the system prompt, conversation history, other Skills' metadata, and your actual request. Default assumption: Claude is already very smart—only add context Claude doesn't already have.
Supported Platforms
Install skills on your favorite agent framework.
Ready to build?
Check out the official documentation or browse existing skills for inspiration.