Agent Skill
What is an agent skill?
An agent skill is a package of instructions, and optionally scripts, that an AI agent can use to carry out a specific task. Each skill lives in its own folder. The folder's name is the skill's name, and a SKILL.md file inside it defines what the skill does.
Anthropic introduced skills for Claude, and many other agents are now adopting them.
How is an agent skill different from a slash command?
Like a slash command, a skill stores prompts and procedures you want to reuse. But skills differ in three ways:
- An agent can invoke them. You trigger a slash command by typing it. An agent uses a skill when it recognizes the task, or you can ask for one by name.
- They can bundle code. A skill can include scripts, so you can extend what the agent can do with deterministic code.
- They work across tools. Claude and a growing number of other agents support skills, so you can define a skill once and use it in many places.
Where do agent skills live?
Each agent looks for skills in its own folders. Claude Code, for example, stores them globally in ~/.claude/skills/, where every project can use them, or in a project's .claude/skills/ folder. Most syncing services won't share a folder in your user directory across devices or with colleagues without extra setup.
Are third-party agent skills safe?
Skills can, and often do, include third-party code, and bad actors use skills directories to distribute malicious code. Before using someone else's skill, read through its files to see what it does. A safer option is to ask your agent to recreate the skill for you, so no unknown code ever touches your data.
Learn more:
- How to Use Claude Code: A Guide to Slash Commands, Agents, Skills, and Plug-ins
- How to Share Your AI Context and Skills Across Devices
- What I Learned from the Recent Wave of Package Hacks (And Is Cowork Immune?)
Related terms:
Last Updated: November 17, 2026