Anthropic’s new ‘Teach Claude a Skill’ feature lets AI agents learn by watching your screen

Anthropic just made training an AI agent feel less like programming and more like onboarding a new hire. The company launched “Teach Claude a Skill” on July 21, allowing users to record their screen, narrate what they’re doing, and turn the whole thing into a reusable skill that Claude can execute on its own later.

How it actually works

The feature lives inside Claude’s desktop app under a “Record a Skill” option, available through the Claude Cowork interface. Users on Pro, Max, and Team plans can access it. Free-tier subscribers are left out.

Here’s the setup: you hit record, perform a task on your desktop, and narrate what you’re doing as you go. Claude captures screen activity, clicks, typing, and voice prompts. It then packages all of that into a skill it can replay autonomously in future sessions.

There are limitations worth noting. Pro plan users face a cooldown period of five hours between recordings. Early user feedback suggests that skill creation is fairly token-intensive, meaning you might bump into rate limits faster than expected.

Why crypto builders care about a productivity tool

Anthropic’s official launch materials make zero mention of crypto tokens or blockchain technology. But the crypto and Web3 communities are already running with it.

Users have begun creating skills for tasks like smart contract auditing and on-chain research. If you’ve ever spent an afternoon manually cross-referencing contract addresses across block explorers, checking function calls, and verifying deployment parameters, you understand why someone would want to teach an AI to do that instead.

A security researcher could record themselves walking through a smart contract audit process: pulling up the code on Etherscan, checking for common vulnerability patterns, cross-referencing with known exploit databases, and documenting findings. Claude captures the entire sequence and can then replicate it across different contracts.

For on-chain analysts, the use case is similarly compelling. Recording a research workflow that involves checking wallet movements, tracking token flows across protocols, and compiling data into a standardized format could save hours of repetitive work. The demonstration-based approach means you don’t need to write elaborate custom instructions or build a separate automation pipeline.

The competitive landscape and what investors should watch

The risk side deserves attention. High token consumption for skill creation means costs can add up quickly, especially for power users running complex multi-step workflows. If Anthropic tightens rate limits or raises prices on the plans that support this feature, it could dampen adoption among the independent developers and small teams that make up a significant portion of the crypto builder community.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.