Anthropic shipped a feature called Record a Skill. You record your screen doing a task once, you talk through it while you work, and Claude turns the whole thing into a skill it can run again. No prompt engineering. No SOP nobody reads. This guide covers the setup, the limits nobody mentions, exactly what to say out loud, and the mistakes that ruin a recording.
Verified against Anthropic support documentation, September 2026
Record a Skill lives inside Cowork in the Claude desktop app. Claude captures your screen, your clicks, your typing and your voice, then reviews the session and proposes a skill. You approve it, and it goes into your skill library where Claude loads it again when the work matches.
The important part is the voice. The shortcuts, the exceptions and the "we never do it this way" parts never make it into a written SOP, and they are exactly what you say out loud without thinking while you work.
| THE FACTS | AS OF SEPTEMBER 2026 |
|---|---|
| Where it lives | Cowork, in the Claude desktop app for Mac. Not in chat, not on Windows. |
| Who gets it | Pro, Max and Team plans. Not Free, not Enterprise. |
| Recording length | About ten minutes. A countdown warns you at roughly one minute left, then it stops on its own. |
| What it captures | Screen, mouse, keyboard and voice narration. |
| What is kept | The video and audio are not retained. A set of screenshots stays in the Cowork task and is deleted when you delete the task. |
Claude needs Screen Recording to see your screen and Accessibility to follow your mouse and keyboard. macOS may make you restart Claude after you grant them. Do this before you plan to record, not while a colleague waits on a call.
In a Cowork task, click the plus button in the composer and choose Record a skill. There is a second door: Customize > Skills > Add > Record your screen. Pick your microphone, then click Start recording.
Not the clean version. The real one. If you always check a second tab before you approve something, check it. If you skip a field on purpose, skip it. A demonstration of an idealised process teaches Claude a process nobody actually runs.
This is the difference between a skill that works and a skill that copies your clicks. Say why, not what. See the next page for the four things worth narrating.
Claude asks about the gaps in your demonstration. Answer them properly, that is where the edge cases get in. Then open the skill it proposes and read it before you save. Check the conditions it thinks trigger the skill and the exclusions it wrote down, because it is generalising from one example and one example is a small sample.
Narration is the whole feature. Four categories carry most of the value, and all four are the things that live in your head and never in the written process.
| RECORD THIS | NOT THIS, YET |
|---|---|
| The task you do every Monday | Anything you have done twice |
| The process with a consistent shape | Work that changes every time |
| The one full of your judgment calls | Anything touching passwords or client secrets |
| The thing you keep re explaining to a new hire | A task that takes 40 minutes end to end |
Then run it on something new before you trust it. Give it a case it has not seen and watch the judgment calls specifically, because that is where a single demonstration generalises badly. Fix the skill, run it again, and only then let it run without you watching.
All trademarks belong to their owners. Easy Hive is not affiliated with or endorsed by them. This site is not part of Facebook, Instagram or Meta Platforms, Inc., and is not endorsed by Meta. Savings and results shown are examples, not guarantees; yours will vary. Educational content only, not legal, tax or financial advice. Full disclaimers.