What are Claude Skills?
A Claude Skill is a folder with a SKILL.md file inside it, plus any scripts, templates or reference files the task needs. Anthropic's platform docs describe skills as "reusable, filesystem-based resources that give Claude domain-specific expertise," and contrast them with prompts, which are "conversation-level instructions for one-off tasks." You write the procedure once. Claude reads it only when a request fits.

The loading is staged, which is why a library of skills stays cheap. Anthropic's docs call it progressive disclosure and give three levels. At startup Claude sees only each skill's name and description, about 100 tokens per skill. When your request matches a description, Claude reads the body of SKILL.md, under 5,000 tokens. Bundled files and scripts load only if the instructions point to them, and a script's code never enters the context: only its output does.
That makes the description the part that matters most. The docs say the description "is what Claude matches your request against when determining whether to trigger the Skill," so it has to say what the skill does and when to use it. A vague description means the skill never fires.
Skills differ from two things people confuse them with. Anthropic's help center says projects "provide static background knowledge that's always loaded," while skills "provide specialized procedures that activate dynamically when needed." And MCP connects Claude to outside services, while a skill teaches Claude how to do a task well. The two work together. What is MCP covers the connection side, and Claude Projects covers the always-loaded side.
Which plans get Claude Skills?
Every Claude plan has skills, and Anthropic's help center says Free accounts can upload their own. The help article "Use skills in Claude," read on 15 November 2026, says that on Free, Pro, Max and Ultra "you can enable example skills and upload your own in Customize > Skills." It needs code execution and file creation switched on. Anthropic's platform docs still list custom skills for "Pro, Max, Ultra, Team, and Enterprise plans," so the two pages disagree. The help center covers the apps themselves, so treat it as the newer word for Free. Check Customize > Skills on your own account.
The built-in skills cover four file types: Excel spreadsheets, Word documents, PowerPoint presentations and PDFs. Claude uses them without setup whenever you ask for one of those files. Anthropic's pre-built skill IDs on the API are pptx, xlsx, docx and pdf.
| Surface | Custom skill goes in | How you add it | Network access | Who sees it |
|---|---|---|---|---|
| Claude apps (claude.ai) | Customize > Skills | Upload a ZIP of the skill folder | Full, partial or none, set by user or admin | You, until you share it (Team and Enterprise) |
| Claude Code | ~/.claude/skills/ or .claude/skills/, plus a synced copy of your claude.ai skills | Create the folder, or sign in with the same claude.ai account | Same as any program on your computer | You, or your repo, or a plugin's users |
| Claude API | The Skills API, /v1/skills | Upload, then list the skill ID in container | None, and no package installs | Everyone in the workspace |
The table is compiled from Anthropic's platform docs and help center on 15 November 2026. Team and Enterprise owners get one more control. They turn skills on under Organization settings > Plugins & skills, on the Policy tab, and shared skills are view-only for the people who receive them. Sharing to groups is Enterprise only.
How do you turn on and use a skill in Claude?
You turn skills on in settings, then switch each skill on under Customize > Skills. These steps follow Anthropic's help center as of November 2026. The file switch is called "Code execution and file creation," and it sits under Settings > Capabilities on individual plans.
- Open Settings > Capabilities and turn on code execution and file creation.
- Open Customize > Skills.
- Toggle the skills you want on. The built-in ones are already working when you create documents.
- Ask for the task in plain words, for example "Turn these notes into a five-slide deck." Claude decides whether a skill applies, so you don't name it.
- Open Claude's thinking on the reply if you want proof. Anthropic suggests checking it to confirm the skill loaded.
To add your own, package the folder and upload it. In Customize > Skills, click "+", then "+ Create skill", choose "Upload a skill" and pick the ZIP. Two rules cause most failures. The ZIP must contain the skill folder as its root, not loose files. And the folder name must match the skill name. Anthropic's listed errors are a missing skill.md file, a folder name that doesn't match, and a ZIP over the size limit, though the page gives no number for that limit. Note that the help page spells the file skill.md while the platform docs spell it SKILL.md. Use SKILL.md.
On a Mac you can skip the writing. Anthropic says you can record your screen in Cowork in Claude for Mac and Claude proposes a skill for you to review. Recording isn't available in chat, on Windows, or on Free and Enterprise plans.
How do you create a Claude Skill from scratch?
You create one by writing a SKILL.md with a name, a description and the steps. The file starts with YAML frontmatter, and name and description are the two required fields. Here is a small example for a weekly status report. It is ours, written for this page, not one of Anthropic's.
`---
name: weekly-status-report
description: Turns raw notes into a one-page weekly status report. Use when the user pastes meeting notes, task lists or a changelog and asks for a status update or weekly report.
---
# Weekly status report
- Group the notes under Done, In progress and Blocked.
- Write one line per item, starting with a verb.
- Put anything blocked first, with who can unblock it.
- End with the three priorities for next week.
Keep it under 250 words. If a date is missing, ask for it. Never invent one.`
Put it in a folder called weekly-status-report, ZIP the folder, and upload it as above. Anthropic's field rules, from the platform docs: name is at most 64 characters, lowercase letters, numbers and hyphens only, and may not contain the reserved words anthropic or claude. So a skill called "claude-helper" is rejected. The platform docs allow a description of up to 1,024 characters. The help center article says 200, so write 200 or fewer and it works everywhere.
Three habits from Anthropic's help center are worth copying. Make one skill per workflow, not one giant skill. Start with plain Markdown instructions before adding scripts. And include an example input and output, so Claude sees what finished looks like. Never put API keys or passwords in the folder.
Anthropic's own examples of custom skills give you ideas for what to build. They include applying brand style guidelines to documents and presentations, generating communications that follow company email templates, and creating tasks in tools like JIRA, Asana or Linear in line with team conventions. A fourth that fits almost anyone: a skill holding your house format for a recurring document, the way the example above does.
How do Claude Code skills and API skills differ?
Claude Code skills live in plain folders and need no upload, and API skills are uploaded objects with IDs. Both read the same SKILL.md format. Claude Code also pulls in your claude.ai skills, but the API does not.
In Claude Code, the docs give these locations. A personal skill is ~/.claude/skills/<skill-name>/SKILL.md and loads in all your projects on that machine. A project skill is .claude/skills/<skill-name>/SKILL.md and loads for sessions in that repository, and the docs advise committing it so your team gets it. A plugin can also carry skills, which then appear as /plugin-name:skill-name. You run one by typing /skill-name, or let Claude load it when it fits. Since version 2.1.273, signing in with the same claude.ai account downloads your enabled skills into ~/.claude/skills/synced/, listed under "claude.ai sync" in /skills. Claude Code never uploads, so edit synced skills on claude.ai. Set syncClaudeAiSkills to false to stop it. How to use Claude Code covers the install.
Claude Code accepts more frontmatter than the apps do. Its docs list fields such as disable-model-invocation (set it to true and only you can trigger the skill), allowed-tools, model and paths, and say all fields are optional with only description recommended. Two limits to know: keep SKILL.md under 500 lines, and the combined description and when_to_use text is truncated at 1,536 characters in the skill listing. This small example comes from the Claude Code docs.
`---
description: Summarizes uncommitted changes and flags anything risky. Use when the user asks what changed, wants a commit message, or asks to review their diff.
---
## Current changes
!git diff HEAD
## Instructions
Summarize the changes above in two or three bullet points, then list any risks you notice such as missing error handling, hardcoded values, or tests that need updating. If the diff is empty, say there are no uncommitted changes.`
On the API, you list skills in the container parameter of a Messages request and include the code execution tool. A request names a pre-built skill like this, using the shape in Anthropic's skills guide: "container": { "skills": [ { "type": "anthropic", "skill_id": "pptx", "version": "latest" } ] }. For your own, you upload with POST https://api.anthropic.com/v1/skills, then pass the returned ID with "type": "custom". The guide says you can include up to 20 skills per request and that the total upload must be under 30 MB uncompressed. API skills run with no network access and can't install packages.
One catch for businesses. Anthropic says Agent Skills isn't covered by zero data retention arrangements, so skill definitions and execution data follow its standard retention policy.
Are Claude Skills safe, and what can't they do?
A skill is code and instructions you hand to Claude, so treat it like software you install. Anthropic's docs say it directly: "Use Skills only from trusted sources: those you created yourself or obtained from Anthropic." A malicious skill can steer Claude into running code or calling tools in ways that don't match what it claims to do. Open every file in a downloaded skill before you enable it, and look hardest at anything that fetches from a URL.
They also have plain limits. The API doesn't sync with claude.ai or Claude Code, so an API skill is its own copy. Claude Code syncs only when you sign in with a claude.ai account, and it never sends edits back. Custom skills don't move when you migrate a personal account to a Team or Enterprise organization, so download them first. And a skill is guidance, not a guarantee: Claude chooses whether to load it from the description, so a skill you expected to fire sometimes doesn't. Test with several prompts after each change.
Do ChatGPT and Gemini have skills too?
Yes, both added them, and the three differ on where you can build one. The table is compiled from each company's help pages as of 15 November 2026.
| Claude | ChatGPT | Gemini | |
|---|---|---|---|
| Format | Folder with SKILL.md | Folder with SKILL.md | Instructions, optionally a folder with SKILL.md |
| Where you build one | Customize > Skills, Claude Code, API | Desktop app, Codex CLI, IDE extension; bundled in plugins on web and mobile | gemini.google.com sidebar, Customize > Skills |
| How you run it | Automatic, matched by description | Type @ to pick one, or automatic | Type / and the name |
| Plans | Free to Enterprise listed; custom upload on Pro and up | OpenAI's skills page names no plan | All Google AI tiers, users 18 and over |
| Replaces | Nothing | Nothing | Gems: read only on 17 November 2026, removed 17 December 2026 on personal accounts |
Gemini is the one moving fastest. Google announced skills in Gemini chat on 30 September 2026 and says "Gems will go away on personal Google accounts" in favor of them. Skills don't yet support the default Gem tools: Create video, Create music, Canvas, Deep research and Guided learning. Workspace business accounts follow in March 2027 and education accounts in June 2027. Our ChatGPT vs Claude vs Gemini guide covers the rest of the three-way comparison.
ChatGPT's help page for skills says standalone skills are available in the ChatGPT desktop app, Codex CLI and IDE extension, and that skills bundled in plugins reach Chat and Work on the web, desktop and mobile. OpenAI's page also names a built-in creator you reach with @skill-creator in Work. It names no plan, so check your own account before you count on it.
What if you want skills and more than one model?
Skills are tied to one vendor's app, so a team that uses Claude, ChatGPT and Gemini keeps three sets. If you mainly want to try Claude beside GPT and Gemini without three subscriptions, Whizi Pro carries all three families in one app, and its first week is $1.99. The free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5, with no card.
Be clear about what that is and isn't. Whizi doesn't run Anthropic's Agent Skills, so a SKILL.md ZIP does not upload there. It has its own feature, also called Skills: written instructions you switch on for one chat or for every chat. They hold text, not scripts or bundled files. For anything that needs Claude's document skills, the Claude app is the right place, and it's the one to pay for if skills are the main reason you're here.
- Turn on code execution and file creation in Settings > Capabilities before you look for the Skills page
- Write the description as what the skill does plus when to use it, and keep it to 200 characters or fewer so it passes in the apps
- Name the folder exactly like the skill name, and ZIP the folder itself, not its loose files
- Keep names to lowercase letters, numbers and hyphens, and leave "claude" and "anthropic" out of them
- Open every file in a skill you didn't write before you enable it
- Plan for the API as a separate copy: a skill uploaded to claude.ai reaches Claude Code on sign-in (2.1.273 or later) but not the API
Frequently asked questions
Are Claude Skills free?
Yes. Anthropic's help center lists skills on Free, Pro, Max, Ultra, Team and Enterprise, and says Free users can enable example skills and upload their own. Its platform docs still list custom uploads as Pro and above, so a Free upload option may lag. Claude Code skills are plain folders on your own machine.
How do I add skills to Claude?
Open Settings > Capabilities, turn on code execution and file creation, then go to Customize > Skills. Toggle a built-in skill on, or click "+", choose "+ Create skill" and upload a ZIP of your skill folder. In Claude Code you skip the upload and put the folder in ~/.claude/skills/ or .claude/skills/.
Where do Claude Code skills go?
Personal skills go in ~/.claude/skills/<skill-name>/SKILL.md and load in all your projects on that machine. Project skills go in .claude/skills/<skill-name>/SKILL.md and load for sessions in that repository. Commit the project folder and your team gets the skill too. Plugins can carry skills as well.
Do Claude Skills work in the API?
Yes. List the skill IDs in the container parameter of a Messages request and include the code execution tool. You can use the pre-built pptx, xlsx, docx and pdf skills or upload your own through /v1/skills. Anthropic's guide allows up to 20 skills per request, and API skills run without network access.