What are agent skills?
A plain-English guide to SKILL.md folders: what is inside one, how an agent decides to use it, and when a skill beats a long prompt.
Updated Oct 22, 2026 · 4 min read
An agent skill is a folder that teaches an AI agent one repeatable job. At its center is a SKILL.md file: a short description at the top, then instructions the agent follows. Next to it sit whatever the job needs, such as scripts, templates, fonts or example files.
What is inside a skill folder
kinetic-title-drop/
├── SKILL.md # description + step-by-step workflow
├── templates/ # HTML compositions the agent edits
├── scripts/render.mjs
└── LICENSES/ # licenses of any open-source partsThe description at the top of SKILL.md is what the agent reads first. When your request matches it, the agent loads the rest of the file and follows the workflow. That keeps your context small: the agent only reads a skill when it needs it.
Why use a skill instead of a long prompt
- It is repeatable. The same skill gives the same look next week and for a teammate.
- It can ship files. Templates, fonts and render scripts travel with the instructions.
- It can check its own work. A good skill ends with a QA step, such as checking that text fits inside safe areas.
Which agents can use skills
Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI and OpenCode can all work from SKILL.md folders. Claude Code reads ~/.claude/skills. The others also read a shared ~/.agents/skills folder. The Lookbook installer writes to both, so one install covers every agent on your machine.
Read next
How to install a skill in Claude Code
Three ways to add a skill to Claude Code: paste the install prompt, run the command, or unzip it into your skills folder by hand.
Which open-source licenses let you sell a skill?
MIT, Apache 2.0, GPL, AGPL and source-available licenses, explained for people who package and sell agent skills.