--- id: claude-skills lang: en title: "Claude Skills" seoTitle: "What are Claude Skills?" description: "Claude Skills are packaged instructions that do a set task the same way every time. What they are, when to use one, and how to build your own custom Skill." part: working-smart partTitle: "Working smart" partOrder: 4 order: 2 level: Intermediate durationMinutes: 6 outcome: "Know what a Skill is and spot the repeatable tasks worth turning into one." summary: "A Skill is packaged instructions that make Claude do a repeatable task the same way every time. Spot the tasks you redo often and turn them into one." draft: false updated: "2026-07-26" furtherReading: - title: "What are Skills?" url: "https://support.claude.com/en/articles/12512176-what-are-skills" note: "Anthropic's plain-language intro in the help centre." - title: "Using Skills in Claude" url: "https://support.claude.com/en/articles/12512180-using-skills-in-claude" note: "How Skills switch on and fire in claude.ai, step by step." - title: "How to create custom Skills" url: "https://support.claude.com/en/articles/12512198-creating-custom-skills" note: "The exact steps to build and upload your own." - title: "Equipping agents for the real world with Agent Skills" url: "https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills" note: "Anthropic on the idea behind Skills and where they shine." keywords: - Claude Skills - what are Claude Skills - custom Claude Skill - reusable AI instructions - automate tasks with Claude quiz: intro: "Five quick questions. Ace them all and you've got Skills down." passScore: 5 questions: - q: "What is a Claude Skill, in plain terms?" options: - "A faster model you pay extra for" - "A packaged set of instructions (and sometimes templates or scripts) that tells Claude how to do a specific task the same way every time" - "A limit on how many messages you can send" - "A separate app you install on your phone" answer: 1 explain: "A Skill is a small bundle: instructions for one job, plus any templates or scripts it needs. Teach Claude the job once, and it does it your way every time." - q: "How do you get a Skill to run on a task?" options: - "Type its exact name as a command every time" - "Just describe the task in plain language; Claude matches it to a Skill and uses it automatically" - "Email Anthropic to switch it on" - "Restart the chat" answer: 1 explain: "Skills are model-invoked. You describe what you want, and Claude picks the matching Skill based on its description. You never have to call it by name." - q: "Why can you keep many Skills around without slowing Claude down or filling the window?" options: - "Skills are stored on your phone, not in the chat" - "Claude only keeps each Skill's name and one-line description in mind, and opens the full instructions only when a Skill actually matches your request" - "Skills delete themselves after one use" - "They don't use any text at all, ever" answer: 1 explain: "This is the clever part, and it ties straight back to lesson 4.1. Until a Skill is triggered, only its name and short description sit in context (a tiny cost). The heavy instructions load on demand." - q: "Which task is the best fit for a Skill?" options: - "A one-off brainstorm with no set format" - "A repeatable job with a clear 'right way', like formatting every report in your house style" - "Asking a single factual question" - "A casual chat about your weekend" answer: 1 explain: "Skills earn their keep on repeatable, well-defined tasks done the same way each time. For a one-off or open exploration, a plain prompt is the better tool." - q: "You find a Skill online from someone you've never heard of. What's the smart move?" options: - "Install it right away; all Skills are safe" - "Be cautious: only use Skills you made or got from a trusted source, and audit anything else first, because a Skill can run code and use tools" - "Forward it to your whole team immediately" - "Skills can't do anything risky, so it doesn't matter" answer: 1 explain: "A Skill can direct Claude to run code and use tools, so a bad one is a real risk. Stick to Skills you built or got from Anthropic, and review anything from elsewhere before you trust it." pass: badge: "Skills unlocked" title: "You've turned repetition into a reflex." body: "Full marks. You know what a Skill is, when it fires, why it stays cheap, and where it fits. Next: Projects, where your files, instructions, and chats live together for ongoing work." retry: "Check the marked answers and give it another go. You've got this." --- > **Level:** Intermediate ยท **Reading time:** ~6 min > > **What you'll be able to do:** know what a Skill is and spot the repeatable tasks worth turning into one. *(Last checked: July 2026.)* Notice how often you tell Claude the same thing? "Use our house style. Headline, then three bullets. British spelling. Sign off as the team, not me." Every time, from scratch. A **Skill** is how you stop repeating yourself. **Teach Claude a job once, and it does that job your way every time.** ## What a Skill is A Skill is a small **package of instructions** for one specific task. At its heart is a short file that says two things: *what* the task is, and *how* to do it, step by step. It can also carry extras the job needs: a template to fill, a checklist to follow, a bit of script to run. The best way to picture it: a Skill is an **onboarding note for a new hire**, but for a single job. Not "here's everything about the company," just "here is exactly how we format a client report, every time." Hand that note over once, and the new hire never has to ask again. ## Two kinds you'll meet **Ready-made Skills.** Anthropic ships a set for everyday document work: Word, Excel, PowerPoint, and PDF. These switch on by themselves. Ask Claude to "build me a slide deck" and the PowerPoint Skill quietly does the heavy lifting. You have probably used one already without knowing its name. **Your own Skills.** This is where it gets useful. You package a task *you* repeat: "format any report in our brand style," "turn my rough notes into our standard client update," "run this checklist over a draft contract." Once it is added, Claude reaches for it whenever a request fits. ## The clever part: Skills stay cheap Remember the context window from lesson 4.1, the desk that fills up? You would think a pile of Skills would clog it. They don't, and here is why. Claude keeps only each Skill's **name and one-line description** in mind, a tiny cost, a handful of words each. It reads the *full* instructions only when a request actually matches. The rest of the time they sit on the shelf, costing almost nothing.
Skills sit on a shelf until one matches Several Skills are shown as cards on a shelf, each showing only a name and a one-line description. When a request matches one, that Skill's full instructions open; the others stay closed and cost almost nothing. On the shelf until needed brand-report Format a report in our style client-update Notes into our standard update contract-check Run our review checklist more... Each costs only its one-line label You: "Turn these notes into a client update." client-update opens Full instructions load, just for this task. The others stay shut. Claude matched your words to the right Skill's description, and reached for it on its own.
The upshot: you can build up a whole toolbox of Skills and pay almost nothing to keep them, because **only the one you need ever opens.** ## You don't summon it, it shows up Here is the part people find surprising. You **do not call a Skill by name.** You just describe the task in plain words, the way you always have. Claude reads your request, compares it to what each Skill says it is for, and reaches for the right one on its own. That puts all the weight on one thing: the Skill's **description**. It has to say clearly *what the Skill does* and *when to use it*, because that description is the trigger. A vague description ("helps with documents") never fires at the right moment. A sharp one ("format a report in our brand style, use when the user asks for a report or a write-up") fires exactly when it should. ## Where Skills earn their keep, and where they don't Reach for a Skill when a task is **repeatable and has a right way to do it.** Formatting to your brand. A weekly report that always takes the same shape. A review checklist you run over every draft. Filling a form the same way each time. These are jobs where consistency is the whole point, and a Skill nails consistency. Skip the Skill for **one-offs and open exploration.** Brainstorming, a single question, a first rough draft of something new: a plain prompt is faster and freer. Do not package a task you will do once. A quick way to tell a Skill apart from things you already know. A **prompt** is a one-off instruction for right now. Your **profile instructions** (lesson 2.5) are broad background that colours every chat. A **Skill** sits in between and is sharper than both: a specific job, packaged, that fires by itself when it fits and can carry its own templates and steps, all without cluttering the window until it runs. ## Making your own You do not need to be technical. A Skill is really just a short text file, plus any templates you want it to use.
The anatomy of a Skill A Skill file has three parts: a name, a description that says what it does and when to use it (this is the trigger), and the step-by-step instructions. It can also bundle template or reference files. What's inside a Skill Name A short label, like brand-report. Description What it does + when to use it. This line is the trigger. Make it sharp. Instructions The steps Claude follows, in order. Optional: bundle a template, a checklist, or a reference file the job needs. In claude.ai you upload it in Settings (under Capabilities), on paid plans with file creation switched on. Each Skill stays private to your own account.
So the recipe is: write down the **name**, a crisp **description** (what plus when), and the **steps** in order. Add a template if the job needs one. Then, in claude.ai, upload it in your **Settings** (look under Capabilities), which works on the paid plans with file creation switched on. From then on, Claude uses it whenever a matching task comes up. One note for teams: a custom Skill you upload stays with *your* account, it is not shared across the whole organisation, so each colleague adds their own copy. The help-centre links below have the exact click-by-click steps. ## One safety note A Skill can tell Claude to run code and use tools, which is exactly what makes it powerful, and also what makes a bad one dangerous. So keep the habit from lesson 1.4: **only use Skills you built yourself or got from a trusted source** like Anthropic. If a Skill turns up from somewhere you do not know, treat it like software from a stranger: read it first, or leave it alone. ## Where this goes next A Skill is the *how*, a job done your way on demand. Next we cover the *where*. **Projects** (lesson 4.3) keep your files, your instructions, and all the related chats together in one place, so ongoing work has a home. Skills and Projects are a natural pair, and together they turn Claude from a clever stranger into a colleague who knows your work. ## Try it Write down three tasks you find yourself re-explaining to Claude, the ones where you always paste the same setup. Now pick the one with the clearest "right way" to do it. Sketch its Skill in five lines: a name, a one-line description (what it does and when to use it), and three or four steps Claude should follow. You do not have to build it yet. Just seeing a real task from your week break cleanly into those parts is the point, that is the moment Skills stop being abstract. *(General exercise. Role and industry versions come once accounts are in.)* ## Key terms [Skill](/glossary#skill), [Claude](/glossary#claude), [Context window](/glossary#context-window), [Prompt](/glossary#prompt), [Project](/glossary#project)