How to Create a Claude Skill

If you keep giving Claude the same instructions, for the same report format, the same review checklist, or the same multi-step procedure, you can stop repeating yourself. A skill packages those instructions once, and Claude follows them every time a task calls for it. Writing one takes less than you might think: a folder, one text file, and a clear description. Here is how to create a Claude skill, and how to make sure Claude actually uses it.

What a skill is

A skill is a set of instructions that Claude loads when a task matches it. Anthropic already ships a few: when you ask Claude for a Word document, a spreadsheet, a presentation, or a PDF, it follows a built-in skill for that file type without you noticing. Your own skills work the same way, except the instructions are yours.

A skill is not the right tool for everything. A one-off request belongs in the conversation. Instructions you want applied to every chat belong in your personal or project instructions. And live data from another app comes from a connector, not a skill. A skill is for a task you repeat and want done the same way each time.

How Claude decides to use a skill

Claude does not read every skill at the start of a conversation. At first, it only sees each skill's name and a one-line description. When your request matches a description, Claude loads that skill's full instructions, and it opens any extra files those instructions mention only at the step where they are needed.

That is why the description matters more than anything else in a skill. It is the only part Claude sees before it decides whether to use the skill at all.

What goes inside a skill

A skill is a folder named after the skill, and the only file it must contain is SKILL.md. The file starts with a short header that holds two fields: a name and a description. The name uses lowercase letters, numbers, and hyphens, and matches the folder name, for example meeting-notes. Below the header come the instructions, written as ordinary text.

Write the instructions the way you would brief a new colleague: the steps in order, an example of a good result, the format you expect, and the cases that tend to go wrong. When a skill needs more than that, the folder can also hold reference documents, templates, and scripts, and the instructions name each file at the step where Claude should use it. Plenty of useful skills never need any of this.

Writing a description that works

A good description says two things: what the skill does, and when to use it. Be specific about the first. "Helps with notes" is too vague. "Turns raw meeting notes into a summary, a list of decisions, and action items with owners" tells Claude exactly what the skill is for.

For the second, use the words people actually type. If your team asks for minutes, a meeting recap, or action items, put those words in the description, because that is what Claude matches requests against. A useful habit is to write the task down in one sentence, along with what a good result looks like. That sentence often becomes the description.

Uploading and testing the skill

Before you upload anything, check that code execution is turned on, because skills run in Claude's code execution environment. You will find it in Settings, under Capabilities, as Cloud code execution and file creation. On Team and Enterprise plans, an Owner controls this setting for the whole organization.

Then zip the skill folder itself, not the files inside it. Claude looks for the folder at the top of the archive, with SKILL.md inside it, so a SKILL.md sitting loose at the top of the ZIP file is not recognized. Upload the ZIP file from Settings, under Skills, and turn the skill on.

Now test it. Send a few requests that should trigger the skill, worded the way people really ask, and check Claude's thinking to confirm it loaded the skill. If it does not pick the skill up, look at the description first: make it more specific, add the words your requests use, and try again. When you open a skill, its files open beside the conversation, so you can edit them directly or have Claude revise them with you.

Letting Claude write the skill

You do not have to write every skill by hand. Describe the task to Claude in a conversation and ask it to turn that into a skill, and Claude drafts the name, the description, and the instructions for you to review and save. Anthropic also provides a skill called skill-creator, which helps you write a skill and test whether it actually improves Claude's results.

Knowing the structure still pays off. It lets you check what Claude wrote before you save it, and when a skill does not load when it should, you know exactly where to look.

Only turn on skills you trust

A skill's instructions, and any scripts it carries, run as part of your conversation. Anthropic does not review skills you upload yourself or that someone shares with you, and a malicious skill could try to steer Claude into actions you did not intend, or leak your data. Before you turn on a skill from anyone else, open it and read what it does.

Conclusion

Creating a Claude skill comes down to a folder, a SKILL.md file, and a description that says what the skill does and when to use it. Zip the folder, upload it, test it with real requests, and sharpen the description until Claude picks it up reliably. Do that for the tasks you repeat most, and Claude starts doing them your way without being told.

Relevant Training

Writing, uploading, and testing skills is covered step by step in Customizing and Automating Claude, part of Cybrec's Claude AI training program, together with connectors, plugins, and scheduled tasks.