Claude Skills: What They Are and How to Make Your Own
Garrett Pierson
A Claude Skill is a folder of instructions Claude reads only when it needs them — so you teach Claude your way of doing something once instead of re-explaining it every chat. Anthropic’s definition is “folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks.”
If you’ve ever pasted the same three paragraphs of context into a fresh Claude conversation for the fourth time this week, that’s the problem Skills solve.
Where Skills came from
Anthropic launched Agent Skills on October 16, 2025, and expanded them on December 18, 2025 with organization-wide management, a skills directory, and an open standard. The spec now lives at agentskills.io, with Anthropic’s own skills published openly in the anthropics/skills repository — many under Apache 2.0, with the document-creation ones source-available.
The customers Anthropic quoted at launch were doing office work, not programming. Rakuten’s line: “What once took a day, we can now accomplish in an hour.”
How Skills work
A Skill is a folder with a SKILL.md file inside it. That file has two required fields at the top — a name and a description — followed by your instructions in plain markdown.
The clever part is when Claude reads what. Anthropic calls it progressive disclosure, and it loads in three stages:
| Stage | When Claude reads it | What it costs |
|---|---|---|
| Name + description | Always, at startup | ~100 tokens per Skill |
| Your instructions | Only when the Skill is triggered | Under 5k tokens |
| Bundled files and scripts | Only when referenced | Nothing until opened |
So you can have twenty Skills installed and pay almost nothing for the nineteen you aren’t using right now. The description field is what Claude matches your request against, which is why Anthropic says it has to state both what the Skill does and when to use it.
That’s also the difference between a Skill and the project instructions you may already have. Instructions load every single time. A Skill’s body sits on disk until the moment it’s relevant.
What comes built in
Four pre-built Skills ship with the Claude apps and the API: PowerPoint, Excel, Word, and PDF. Claude uses them on its own when you ask for a document — no setup. (These four are not available in Claude Code.)
Beyond that there are partner Skills from Notion, Figma, and Atlassian, plus whatever you or your team upload. Team and Enterprise accounts can provision Skills across the whole organization; on personal plans, each Skill is yours alone.
How to turn Skills on
Two settings:
- Go to Settings → Capabilities and enable Code execution and file creation. Skills won’t run without it.
- Go to Customize → Skills and toggle on the ones you want.
Anthropic’s help center lists Skills as available on Free, Pro, Max, Team, and Enterprise plans. On Team and Enterprise accounts the switches live in Organization settings instead, so if the Skills menu isn’t showing up, that’s the first place to look. We break down the paid tiers in Claude Code pricing.
How to make your own
You do not have to write code, and you do not have to write the file yourself.
Ask Claude to build it. Describe the task, the standards you hold it to, and what a good result looks like. Claude writes the properly formatted Skill file and hands it back. Anthropic’s own tutorial takes this route: “You also don’t have to be technical to start creating skills.”
Record yourself doing it. On the Mac app, Pro, Max, and Team users can hit “Record a skill,” do the task on screen while narrating, and let Claude propose a Skill from the recording. A recording runs about ten minutes, and the video and audio aren’t kept.
Write it by hand. Make a folder, put a SKILL.md in it with a name and description, add any reference files or scripts, zip it, upload it. Naming rules: lowercase letters, numbers, and hyphens only, 64 characters max, and it can’t contain the words “anthropic” or “claude.”
In Claude Code, Skills skip the upload entirely — they’re just folders on your computer. Drop one in ~/.claude/skills/ to use it everywhere, or .claude/skills/ inside a project to keep it with that project. Then call it by typing /your-skill-name. Anthropic merged custom slash commands into Skills, so a command file you already had keeps working exactly as before.
Four Skills worth making if you’re building an app
These are the ones that earn their keep for founders, and each one is a document you already half-wrote in some chat window:
- Your product spec format. Teach Claude how you want a feature written up before any code gets touched — the sections, the level of detail, the acceptance criteria. Our PRD template is a ready-made starting point to paste in.
- Your review checklist. The specific things you make Claude check before you ship: no secret keys in the code, every form validated, the app still builds. Written once, applied every time.
- Your brand voice. Feed it your landing page copy and three emails you like. Now every piece of marketing Claude drafts sounds like you instead of like AI.
- Your launch routine. The steps you take to get a change live, in order, including the ones you keep forgetting. This is the Skill that stops the 11pm deploy mistake.
If you want prompts to build these out of, our 13 Claude Code prompt templates are the raw material — a good prompt you keep reusing is a Skill waiting to be saved.
What Skills are not
Three honest limits, because the marketing around this has been generous:
- They don’t give Claude new abilities. A Skill is instructions and files. It makes Claude reliable at your version of a task; it doesn’t teach the model anything it fundamentally couldn’t do.
- They don’t follow you between apps. A Skill you upload to claude.ai isn’t available in the API, and Claude Code’s Skills are separate from both. Same file, three uploads.
- A Skill from a stranger can hurt you. Anthropic’s warning is blunt: use Skills only from sources you trust, because a malicious one “could lead to data exfiltration, unauthorized system access, or other security risks.” Skills run code. Read every file in one before you install it, the same way you’d read what an app wants access to.
The bottom line
Skills are the boring, useful half of working with AI: you stop re-teaching Claude your standards, and it stops handing back work you have to fix the same way every time. Make the first one out of the instructions you’re already tired of typing, and the second one comes easy.
The tools keep getting better at building. Knowing what to build and how to sell it is the part they can’t do for you, which is exactly what Software Secrets 2.0 covers — and the whole book is free.