What are Claude skills, and how should a team set them up?
A Claude skill is a folder of instructions that teaches Claude how to do one kind of task your way, and Claude loads it only when the task calls for it. Every Claude plan includes skills, and four of them, for Word, Excel, PowerPoint and PDF, already work without setup.
On a Team plan, the owner decides three things first: whether code execution is on, who may publish a skill to the whole company, and which two or three skills everyone gets on day one.
Questions as you read? Ask, or take the context files with you.
Where should we send it?
Sent ✓ Download it now →
You already use Claude skills
Ask Claude for a slide deck and it builds one, with slides you can open in PowerPoint. Ask for a budget in Excel and you get a working spreadsheet with formulas in it. Each time, a skill does that work. Anthropic ships Claude with built-in skills for Word, Excel, PowerPoint and PDF, and in claude.ai Claude uses them “with no setup required,” according to Anthropic’s documentation.
So most teams on Claude are already running skills without having decided anything about them. The decisions start when you want Claude to do a task the way your company does it: your proposal format, your call sheet, your brand voice, your review checklist. That is where custom skills come in, and where an owner has to choose who writes them, who checks them, and who gets them.
On a Team plan, the default lets anyone publish a skill to the whole company without review. That may be fine for your firm. It should be a choice you made.
Your team already runs Claude skills. The decision is which ones to add, and who can publish them to everyone.
What a Claude skill is, in one paragraph
A skill is a small folder with a written description and a set of instructions, sometimes with templates or scripts alongside. Claude reads the short description of every skill it has, and when a task matches, it loads the full instructions and follows them.
Anthropic’s Help Center draws the line against Projects plainly: a Project holds background knowledge “that’s always loaded,” while skills are “task-specific and only load when relevant.” If you want the full anatomy of the file and how to write one that gets used, we cover it in agent skills explained. This article stays with the practical questions for a team on Claude.
The AI Briefing
Tuesdays. 500+ leaders. No hype, just what works.
The skills you already have
The four built-in skills cover the files most firms send every day. Anthropic’s documentation describes them this way:
| Built-in skill | What it does |
|---|---|
| PowerPoint | Creates presentations, edits slides, analyzes presentation content |
| Excel | Creates spreadsheets, analyzes data, builds reports with charts |
| Word | Creates, edits and formats documents |
| Generates formatted PDF documents and reports |
The five context files.
Fill-in templates for what your AI should know, and the page that says where each one loads.
On its way. Check your inbox, or download it now →
These work in claude.ai and the Claude apps as soon as code execution is on, and nobody has to install them. They do not come with Claude Code, Anthropic’s coding tool, which has its own setup.
For a production company or an agency, these four are often the first real win. A treatment drafted straight into a formatted Word file, a budget that arrives as a working spreadsheet, a pitch deck you can open and edit: that removes the copy and paste that eats an afternoon. Before you write any custom skill, check how much of your weekly document work these four already cover.
Which plan, and what a Team plan turns on by default
Anthropic’s Help Center says skills are available on Free, Pro, Max, Team and Enterprise plans. They need code execution switched on. Managing skills for a whole organization is only on Team and Enterprise.
If your firm is on Team, these defaults apply to you. The Help Center describes them for the two business plans:
| Setting | Team plan default | Enterprise plan default |
|---|---|---|
| Owner can give a skill to everyone at once | Yes | Yes |
| Publishing a skill to the whole organization | Open: no review | Requires review |
| Automatic scanning of skills and plugins | Not available | Available |
| Giving skills to specific groups through plugins | Not available | Available |
Publishing is open by default on Team, so a skill one person writes can appear for everyone without anyone checking it. Scanning, Anthropic’s automatic check of a skill before it runs, is an Enterprise feature. On a Team plan, the check is a person, or nobody.
Sharing is loose by default as well. On Team, sharing a skill with a colleague is on, while sharing with a group is off. Custom skills are private to the person who made them until they share them, and the Help Center says admins can see their names and sharing status, “but not their files.” That protects people’s work. It also means an owner cannot read what is inside a personal skill someone on the team is using with client material.
Set up in four decisions
A sensible setup for a Team plan takes four decisions, in this order. The owner or whoever runs your Claude account makes them once, and revisits them when the team grows.
Open publishing suits a five-person studio where everyone talks every day. Once you have people who do not see each other’s work, a single reviewer for company-wide skills is cheap insurance. Sharing and publishing events land in the audit log, according to the Help Center, so you can see who shared what, though the log does not hold the skill’s contents.
For the first provisioned skills, pick the two or three tasks that every person does every week and that have a clear standard in your firm. Those are your first company skills.
Your first custom skill
A task earns a custom skill when three things are true: it repeats, your firm has a standard for it, and more than one person does it. A proposal in your house format, a production schedule in your template, a client recap with your sign-off rules, a brand voice check before anything goes out. A one-off task does not need a skill. Neither does background knowledge Claude should always have; that belongs in a Project.
A skill sits next to three other ways of giving Claude context:
| What it holds | When Claude uses it | |
|---|---|---|
| Project | Background knowledge and files for one body of work | Always, inside that Project |
| Custom instructions | Your general preferences | In every conversation |
| Skill | How to do one kind of task, step by step | Only when the task matches |
| Plugin | A bundle of skills and connectors, packaged for a team | When its skills or connectors are needed |
Anthropic’s Help Center describes two ways to make a skill. You can write one: a folder with a file of instructions, zipped and uploaded under Customize, then Skills. Or, on Pro, Max and Team plans, you can record one in Cowork on a Mac: you record yourself doing the task once, for up to about 10 minutes, and Claude builds a skill from the recording. For people who would never write a set of instructions, recording is the easier way in.
Claude decides whether to load a skill by reading its description, and in claude.ai that description has a limit of 200 characters, according to the Help Center. Write it as a plain sentence about when to use the skill: “Use when drafting a client proposal or a statement of work for a production.” A vague description is the most common reason a skill sits unused.
After you add a skill, ask Claude for the task and watch whether it uses the skill. If it does not, the Help Center’s advice is to check “that the skill’s description field clearly explains when it should be used.” Tighten the description and try again before you rewrite the instructions.
A skill also needs an owner. Name one person who keeps each company skill current, the way someone owns your proposal template today. A skill nobody owns drifts out of date while it keeps getting used, which is worse than having no skill.
A worked example: a call sheet skill
For a production company, the call sheet makes a good first skill: every shoot day needs one, every producer writes it a little differently, and a mistake on it costs a crew hour. The skill is a folder with one file in it, called SKILL.md:
---
name: call-sheet
description: Use when drafting a call sheet for a shoot day from a schedule, a crew list and location notes.
---
Call sheet: house standard
1. If the shoot date, schedule, crew list or location address is missing, ask for it before drafting.
2. Put general call time, location, parking and the nearest hospital at the top.
3. List crew by department in the order we always use: production, camera, grip and electric, art, wardrobe, hair and makeup.
4. Show each person's call time next to their name. Flag anyone whose call is before general call.
5. Add the weather forecast line and the sunset time.
6. End with the producer's name and phone number.
7. Before you finish, check that every crew member on the list appears once, and that every time is in the same format.
Everything in it is plain language. The description is one sentence, well under the 200-character limit, written the way a producer would ask for the task. The instructions are what a good producer already knows and a new one has to learn: what goes at the top, the order of departments, the check before it goes out. Nothing in it is clever. Its value is that the standard now lives in a file the whole team uses, written once by the person who owns it.
A firm might write five or six skills like this in its first months: the call sheet, the treatment format, the client recap, the budget summary, the brand voice check. Each one is a page. The hard part is agreeing what the standard is, which is a conversation your team has with each other.
The five context files.
Fill-in templates for what your AI should know, and the page that says where each one loads.
On its way. Check your inbox, or download it now →
When a skill is not the answer
A custom skill takes work to write and keep current, so it should earn its place.
When the built-in four already do the job, start there. Many firms find that a formatted Word file or a working spreadsheet covers most of what they wanted from a custom skill.
When Claude needs to know about one client or one production, use a Project. A Project holds the brief, the files and the background, and it is always loaded inside that Project. A skill is for a task you do the same way across many clients.
When every instance needs fresh judgment, a skill will not carry it. A skill can hold your checklist for a treatment. It cannot decide whether this treatment is the right idea for this client.
There is also a limit on how many skills help. Anthropic’s guidance for larger deployments warns that with too many skills active, Claude “may fail to select the right Skill or miss relevant ones entirely.” Vercel measured a version of the same problem: in its evaluations, published in January 2026, an agent never used the skill it had available in 56% of test cases. A few well-described skills beat a long list.
The directory, and skills from strangers
Claude now has one directory for skills, connectors and plugins, under Customize. It includes skills from Anthropic and from partners; when Anthropic opened the directory in December 2025, it named partner skills from Notion, Canva, Figma and Atlassian. Directory skills are view-only. To change one, you download a copy and upload it as your own.
Outside the directory, many more skills sit on GitHub and on marketplace sites. Some are excellent. Anthropic’s own advice on all of them is blunt: “Use Skills only from trusted sources: those you created yourself or obtained from Anthropic.” The reason is in the next line of the same document. A skill can carry code and instructions, and Anthropic warns that “malicious Skills could lead to data exfiltration, unauthorized system access,” which means sending your data somewhere it should not go.
For a firm without an Enterprise plan, that leaves three plain rules:
- Read every outside skill before you turn it on, the way you would read a contract.
- Prefer copying the useful part into your own skill over installing someone else’s whole.
- Never install an outside skill on an account that holds client material without one person reviewing it first.
The personal skills people make on their own accounts follow them, which connects to a wider problem we cover in the risks of shadow AI on personal accounts.
What to keep out of a skill
A skill holds how your firm does a task. It should not hold the material of any one job. Keep client names, budgets, contracts and personal details out of the skill files, and let people bring that material into the conversation when they use the skill.
Anthropic’s own pages give the reasons. Admins can see a personal skill’s name but not its files, so client material inside one is invisible to the person responsible for it. Skills are not covered by zero data retention arrangements. And a company skill reaches everyone it is provisioned to, which is the point of a standard and the wrong place for something confidential.
A good test: could you hand the skill file to a new hire on their first day? If yes, it belongs in a skill. If it would need an NDA first, it belongs in the job itself.
Where your skills work
Claude now runs in several places, and skills do not follow you everywhere in the same way. As of October 2026, according to Anthropic’s documentation:
| Where | Your claude.ai skills |
|---|---|
| claude.ai and the Claude apps | Yes, including the built-in four |
| Cowork | Yes; skills carry over as Cowork reaches every conversation |
| Claude in Excel, PowerPoint, Word and Outlook | Yes; Claude applies relevant skills automatically |
| Claude Code | Yes, one way: Claude Code downloads your claude.ai skills and never uploads its own back |
| The Claude API | No; skills there are uploaded and managed separately |
Build your company skills in claude.ai, where the owner can provision them and where they reach the most places. If your firm also builds software with Claude Code, those skills come along. Anthropic’s pages do not say whether skills run in the mobile apps, so do not plan a workflow that depends on it.
Plugins are the newer way to hand out a set of skills. A plugin packages skills with connectors to your other tools, and owners on Team and Enterprise plans can add plugins for the organization and set one as Required, so everyone has it. On Enterprise, owners can target a plugin to a group, such as only the production team.
Keeping skills current
A skill is only as good as the standard it holds, and standards change. When your firm changes how it does a task, the skill has to change the same week, or people get the old way from Claude and the new way from their manager.
Every company skill needs a named owner. The owner reads it again when the house standard changes and at least once a quarter. And anyone who sees Claude get the task wrong tells the owner, with the example.
Claude has started to help with the first step. In September 2026, Anthropic added Smart reports in beta, which suggest “which repeated patterns are worth packaging as shared skills,” according to the Claude release notes. Treat those suggestions as a list of candidates for your team to judge.
When a skill has run its course, switch it off for the company and remove it. If an owner turns Skills off for the organization altogether, the Help Center notes that it also removes the skills that had synced to Claude Code, so check with anyone building software before you do.
Plugins change the scale of this. Anthropic released a Salesforce plugin in beta in September 2026 that comes with “37 pre-built sales skills.” A bundle like that saves setup time, and it also means 37 sets of instructions written for someone else’s sales process. Read them the way you would read any outside skill, and copy what fits.
Common setup mistakes
- Leaving publishing open without deciding to. On Team plans it is the default. Make it a decision.
- A vague description. Claude picks skills from their descriptions. “Proposal helper” will be missed; “Use when drafting a client proposal or statement of work” will be found.
- Too many skills. Each extra skill makes the right one harder to pick. Start with two or three for the whole company.
- No owner. A skill nobody owns keeps running after the standard has changed.
- Installing outside skills on accounts with client material. Read them first, or copy the useful part into your own.
- Expecting skills everywhere. Skills in claude.ai do not reach the Claude API, and Anthropic does not say they run in the mobile apps.
The format behind it
Claude skills follow an open format called Agent Skills. On October 22, 2026, Anthropic announced it is donating the standard to the Agentic AI Foundation, an industry body, and said “more than 70 agent platforms” have adopted it, including ChatGPT, Cursor, GitHub Copilot and VS Code. In the same announcement, Anthropic said one million skills have been published and installed nearly 280 million times, its own figures. That handover is not final yet; the foundation’s records show a board vote still pending.
For an owner, the point is portability. A skill your team writes for Claude is a plain folder of instructions that other tools can increasingly read. The work you put into writing down how your firm does a task is yours, and it moves with you if your tools change. We build our own work on Claude, so weigh our view with that in mind.
A skill is your team’s standard, written down so the tool follows it. The people who wrote the standard still decide when it needs to change. That is what we mean by Humans First AI: the AI follows the rules, and your people own them.
What to do this week
For a firm on a Claude Team plan, the first week looks like this:
- Check that code execution is on, and try the built-in four on this week’s real documents.
- Decide who may publish skills to everyone, and change the setting if Open is not your answer.
- Pick the two or three tasks every person does every week, with a clear house standard.
- Write or record a skill for the first of them, give it a one-sentence description, and name its owner.
- Provision it to the company, and ask the people who do that task most to use it for a week and tell you where it went wrong.
If you are weighing skills against other ways of building with AI, our comparison of AI skills vs AI agents explains why we start with skills. And if you want to see how skills fit into a wider setup, read Claude as an operating system.
Turn One Recurring Task Into Your Team’s First Claude Skill
Paste this into Claude with one real example of the task, such as a recent proposal or call sheet your firm is proud of. You get a draft skill: a description Claude can match on, and step-by-step instructions in your house standard.
Context: I run a [size]-person [type of firm]. I want to turn one task my team does every week into a Claude skill so everyone does it to the same standard. The task is: [task]. Below is one good example of the finished work, and here is what makes it good in our firm: [two or three sentences].
Step 1. Write the description: Write a description of 200 characters or fewer that tells Claude exactly when to use this skill, in plain words a teammate would use when asking for the task.
Step 2. Write the instructions: Write numbered instructions that would let a new hire produce work like my example: the structure, the house rules, what to check before it is done, and what to leave out. Mark anything you had to guess with [CHECK].
Output: The description, the instructions as a file called SKILL.md, and a short list of the [CHECK] items for me to answer.
The prompt drafts the skill. It cannot know the judgment calls your senior people make, so have the person who owns the task read and correct it before you upload it. See where you stand →
Sources
- What are skills? (Claude Help Center, read October 26, 2026): plans, code execution, skills vs Projects and custom instructions
- Use skills in Claude (Claude Help Center, read October 26, 2026): turning skills on, privacy of custom skills, the description check, Office add-ins
- Creating custom skills (Claude Help Center, read October 26, 2026): uploading, recording a skill, reviewing downloaded skills
- Provision and manage skills for your organization (Claude Help Center, read October 26, 2026): provisioning, publishing modes and defaults, sharing, audit log
- Browse skills, connectors, and plugins in one directory (Claude Help Center, read October 26, 2026)
- Manage Cowork plugins for your organization (Claude Help Center, read October 26, 2026)
- Get started with skill and plugin scanning (Claude Help Center, read October 26, 2026): scanning on Enterprise plans
- Claude release notes (Claude Help Center, September 2026): Cowork in every conversation, plugins in the directory
- Agent Skills overview (Anthropic documentation, read October 26, 2026): built-in skills, security guidance, data retention
- Skills in Claude Code (Anthropic documentation, read October 26, 2026): one-way sync from claude.ai
- Donating Agent Skills to the Agentic AI Foundation (Anthropic, October 22, 2026): the standard and the platform count
- AAIF project proposal 47 (Agentic AI Foundation, October 21, 2026): status of the transfer
- Skills for organizations, partners and the ecosystem (Anthropic, December 18, 2025): the directory and partner skills
Frequently Asked Questions
Which Claude plan do we need to use skills with a team?
Anthropic’s Help Center lists skills on Free, Pro, Max, Team and Enterprise plans, but managing skills for a whole organization is only on Team and Enterprise. For a firm of 10 to 100 people, Team covers giving skills to everyone and controlling who can publish them. Enterprise adds automatic scanning and targeting skills to specific groups.
Do we have to install anything to get Claude to make Word, Excel and PowerPoint files?
No. Claude’s built-in skills for Word, Excel, PowerPoint and PDF work in claude.ai with no setup, according to Anthropic’s documentation. Code execution has to be switched on.
How do I add a custom skill, and who in my company can see it?
Zip the skill folder and upload it under Customize, then Skills, or record one in Cowork on a Mac if you are on a Pro, Max or Team plan. A custom skill is private to the person who made it until they share it. Admins can see its name and sharing status but not its files.
How do I give the same skill to everyone in the company?
On Team and Enterprise plans, an owner adds the skill under Organization settings, in Plugins and skills. It reaches everyone at once and is switched on by default, and each person can turn it off for themselves.
Can anyone on our team publish a skill to the whole company?
On a Team plan, yes, by default. The publishing setting starts at Open, which means no review. Owners can change it to Requires review, so one person approves each company-wide skill, or to Off, so only owners provide skills.
Are skills from the Claude directory or GitHub safe to install?
Treat outside skills like software from a stranger. Anthropic advises using skills only from sources you trust, such as ones you wrote or got from Anthropic, and warns that a malicious skill could send your data elsewhere. Read any outside skill before turning it on, especially on accounts that hold client material.
What is the difference between a skill, a Project and a plugin in Claude?
A Project holds background knowledge Claude always uses inside that Project. A skill holds the steps for one kind of task and loads only when the task comes up. A plugin packages several skills with connectors to your other tools, so a team can get them all at once.
Do our Claude skills also work in Claude Code, Cowork and the Office add-ins?
Your claude.ai skills work in Cowork and in Claude for Excel, PowerPoint, Word and Outlook. Claude Code downloads them too, but only in one direction. Skills on the Claude API are managed separately, and Anthropic does not say whether skills run in the mobile apps.
