Explaining it all again, every time?
Your proposals always follow the same structure. The monthly report always has the same columns. Your posts always have the same tone. And every time, you explain it all to Claude from the start, and fix the same mistakes.
A skill closes that gap. You write "this is how we do it here" once, and Claude pulls it out on its own every time the task comes up.
In this lesson: what a skill is, where skills live, and how to build your first one without writing a single line of code.
What a skill is, and how Claude picks one
A skill is a folder with one instructions file called SKILL.md, and sometimes templates, reference files and scripts too. Think of an expert's work binder: when an Excel task comes in, Claude opens the Excel binder.
It doesn't read every skill in every conversation. It sees only the name and the short description of each one, and when your request matches a description, it loads the full instructions. That's why the description is the most important part of a skill: it's the only thing Claude reads before it decides.
Anthropic's built-in skills already work for you: creating and editing Excel, Word, PowerPoint and PDF files. They switch on by themselves when needed. They're the reason that when you asked for a report in Excel in lesson 5, you got a real file and not a table in the chat.
Where skills live
Everything is in Customize > Skills, on the web and in the desktop app. In the Your skills tab they're grouped by source: what you created, what came from your organization, what was shared with you, and what came from Anthropic and its partners. The Discover tab has more.
- Turn a skill on with the Turn on button in its row, or with the toggle at the top of its page.
- Use it by simply describing the task. Want to pick it yourself? Type
/in the message box and choose. - One capability needs to be on: in
Settings > Capabilities, file creation and code execution (in the new interface it's called Create and edit files).
According to the pricing page, skills are available on all plans, as of October 2026. They're saved to your account, and if you also work with Claude Code, you get them there automatically, from version 2.1.273.
The right way to create your own skill
The mistake I see again and again: sitting down to write a skill from scratch, in theory. The way that works:
- Do the task once with Claude, as usual, with all the corrections along the way.
- When the result is exactly what you wanted, ask Claude to turn what you learned into a skill.
- Next time, the task comes out right on the first try.
For the build itself, Anthropic has a dedicated skill called skill-creator, which helps you write skills and test them. Turn it on in Customize > Skills, where it appears with Anthropic's skills. Then, at the end of a task that went well:
"Take everything we did here and turn it into a skill called proposal, so next time I can just type /proposal with the client's name and get a draft in exactly this format. Use skill-creator."
Claude will build the skill file. Save it the way Claude shows you, and if you got a ZIP file, upload it in Customize > Skills: the plus button, Create skill, then Upload a skill.
You can also update it as you go: "Add to the proposal skill that there's always a section on how long the offer is valid." That's it, updated.
Three skills worth building
- Proposal: your fixed template, the pricing rules (before or after VAT, how long it's valid), the tone, and where to save it.
- A post in your voice: three or four real examples, phrases you love and phrases you don't, rules for length and structure.
- Client briefing: what you always check before a meeting, and the format you present it in.
What the three have in common: they hold your "this is how we do it here", the knowledge Claude has no way to guess.
Few skills, and good ones
Anthropic's advice for a good skill, in short:
- Focused. A separate skill for each process. Several focused skills work together better than one giant skill.
- A specific description. When to use it, in the words you actually write when you ask for the task.
- An example of the output. When there's an example of input and output, Claude knows what to aim for.
- A test after every change, on a real request.
When is it not a skill? A one-off task, you just ask for in the chat. Live information from another system is a job for a connector (lesson 10). And instructions that apply to every conversation go into your standing instructions (lesson 7).
Safety. A skill someone shared with you, or one you uploaded from the web, hasn't been checked by Anthropic. Open it, read the SKILL.md, and only then turn it on.
My advice: a skill only for a process that truly repeats. And once a month, go through the list and turn off whatever you haven't used.
Try it now
- Pick a task you've repeated at least twice this month.
- In
Customize > Skills, turn on skill-creator. - Do the task with Claude until the result is spot on, then send:
"We're done. Now turn what we did into a skill. First write me a short description: when to use it, and the words I'll use when I ask for it. Then the instructions, with the example we made today. Show me everything before you save."
- Save the skill and turn it on.
- Open a new conversation, ask for the task in one sentence, and check that Claude uses the skill. If it doesn't, sharpen the description.
What to take with you
A good skill is born from a task that already worked, not from a blank page. And its description is what makes Claude choose it, so it's worth polishing.
