Claude Code skills: what they do and how to build your own
Claude Code skills are folders of instructions the agent loads when a task matches their description: each skill contains a SKILL.md file that tells Claude how to do one specific job, your way. You can let Claude trigger them automatically when relevant, or call one directly by typing /skill-name. It’s the simplest way to turn a procedure you keep re-explaining into know-how installed once and for all — and it doesn’t require coding.
What exactly is a skill?
A skill is a folder containing a SKILL.md file. That file has two parts: a short header with a description — that’s what tells Claude when to use the skill — then instructions in Markdown, plain text: the steps to follow, your rules, the expected output format.
The folder can also carry supporting files: document templates, references, scripts. And that’s where the format is clever: as long as the skill isn’t in use, only its description sits in Claude’s working memory. The full content loads only when a task calls for it.
- A “meeting minutes” skill that turns your raw notes into structured minutes, always in the same format.
- A “review” skill that applies your checklist before any document goes out.
- A “proposal” skill that follows your outline, your standard clauses and your tone.
Skill or CLAUDE.md: which one to use?
Both are written instructions, but they don’t work the same way. The CLAUDE.md (or AGENTS.md) is read at every session: it carries what is always true — who you are, your line of work, your general rules. A skill loads only when needed: it carries a procedure, a piece of know-how.
The rule of thumb from the official documentation is concrete: if you keep pasting the same instructions into chat, or if a section of your CLAUDE.md has grown into a ten-step procedure rather than a fact, it deserves to become a skill.
Where skills live
The folder’s location decides its reach. A skill placed in ~/.claude/skills/ is personal: it follows you across all your projects. A skill placed in .claude/skills/ at the root of a project applies to that project only — and since it’s just a folder of text files, it can be shared with a team like any other project document.
Good to know: Claude Code watches these folders. Adding or editing a skill takes effect within the current session, no restart needed.
Create your first skill, step by step
No special tooling required: a skill is a folder and a text file. The hard part isn’t technical — it’s describing your procedure clearly, the way you’d explain it to someone covering for you.
- Create a folder named after the skill, for example ~/.claude/skills/meeting-minutes/.
- Write a SKILL.md inside: at the top, a description saying when to use it (“Turns raw meeting notes into structured minutes…”); below, the steps and the expected format.
- Test it: type /meeting-minutes in Claude Code, or describe the task and see whether the skill triggers on its own.
- Refine the description if the skill triggers too often or not enough — the description is what drives triggering.
An open format that goes beyond Claude Code
Skills follow the Agent Skills open standard, published by Anthropic (specification at agentskills.io). The same SKILL.md folder is picked up by other agent tools, including OpenAI’s Codex: what you write for Claude Code isn’t locked into one vendor.
You don’t have to start from scratch either: Anthropic maintains a public repository of example skills on GitHub (anthropics/skills) — useful for seeing how good skills are written before drafting your own.
Skills + Context Folder: the two parts of a workspace
A skill describes how to do a task. But to work well, Claude Code also needs to know who it works for: your line of work, your clients, your rules. That’s the Context Folder’s job — an AGENTS.md and supporting files it reads before starting. Skills without context: procedures applied blindly. Context without skills: an assistant that knows you but reinvents the method every time.
The Claude Code AI Workspace Kit prepares that second part for you: you answer a few questions about your line of work, and your Context Folder is generated — the AI’s role, rules, typical missions, examples. The kit is being built: sign up on its page to be told when it ships.
Do I need to know how to code to create a Claude Code skill?
No. A skill is a folder containing a SKILL.md file written in Markdown — plain text. If you can write a clear procedure, you can write a skill. Scripts and automations are possible, but optional.
What is the difference between a skill and a saved prompt?
A prompt gets pasted back into chat by hand, every time. A skill is installed: Claude triggers it on its own when a task matches its description, you can call it with /skill-name, and it can carry supporting files — templates, references, scripts.
Where can I find ready-made skills?
Anthropic maintains a public repository of examples on GitHub (anthropics/skills), and skills are also shared through Claude Code plugins. The most useful ones are often the ones you write yourself: your procedures are where the value is.
Do skills work outside Claude Code?
Yes. The format follows the Agent Skills open standard (agentskills.io), published by Anthropic and adopted by other agent tools, including OpenAI’s Codex. Check each tool’s documentation for support details.
Claude Code AI Workspace Kit
Skills tell Claude Code how to do things. The Context Folder tells it who it works for — AGENTS.md, rules, missions, examples, generated from your line of work. The kit is being built: sign up to be told when it ships.