Tech • AI • Robotics • Game
Claude Learns by Watching You: Train It in 2 Minutes!
7/10TL;DR
Claude’s desktop app includes a screen-recorded skill creation feature that lets users teach recurring tasks by showing and narrating the workflow, then save it as an automated command.
KEY POINTS
A shortcut to building AI skills
The feature sits in the Customization area of the Claude desktop application and offers options such as importing skills, creating one manually, or using Create with Claude and Record a skill. Its purpose is to lower the barrier to creating repeatable AI workflows, especially for users who do not want to write structured prompts or markdown-based instructions from scratch.
Desktop-only capability
The tool is available in the Claude app, not the browser version. It can watch what happens on screen while also capturing spoken instructions, allowing users to demonstrate a process directly instead of describing every step in text.
How the recording works
A user starts recording, shares the screen, and narrates the task while moving through applications and menus. The system observes screenshots, interface changes, cursor activity, and the spoken explanation, then converts that demonstration into a reusable skill that can later be triggered with a slash command.
Example: reviewing YouTube comments
In one test, the feature was used to create a skill that opens YouTube Studio, goes to the Community comments area, switches sorting to most recent, reads the last 10 comments, and produces a short analysis. The requested output included whether comments were supportive or critical, whether criticism targeted the creator or the subject matter, whether any comments suggested improvements, and whether any entries stood out as especially funny or unusual.
Fast turnaround from demo to usable command
The resulting skill was generated in roughly 90 seconds and saved under a custom command similar to /comment-review. Once stored, it could be launched manually in a new conversation with no extra prompt, turning a short demonstration into a persistent workflow.
The system can reason, not just mimic
The generated skill did more than blindly replay instructions. It identified that sorting by relevance would distort a “last 10 comments” review and preferred chronological order. It also noticed YouTube Studio sentiment labels and chose not to rely on them after judging that at least one ironic or friendly remark had been misclassified as negative.
Automation and scheduling
After saving, the skill can be tied to scheduled execution, such as running every day at 9:00 a.m. That would allow the app to deliver a recurring summary a few minutes later, effectively turning a manually repeated check into a daily briefing.
Different ways the task can run
The feature may execute actions through the user’s existing browser, through an internal browser inside the app, or potentially through connected tools and APIs. In the test case, the skill chose to use the already open Chrome session and autonomously navigated to the relevant YouTube Studio pages, clicked menus, sorted comments, and compiled the review.
Read-only behavior for sensitive workflows
The demonstrated setup operated in read-only mode, meaning it reviewed comments without replying, liking, or storing content for other purposes. That distinction matters for users who want help analyzing inboxes, dashboards, or back-office tools without granting the AI permission to take visible public actions.
Broader significance for AI productivity
The feature highlights a larger shift in AI use: practical value increasingly depends not only on the underlying model but on the surrounding workflow, memory, habits, and reusable skills. By letting users teach tasks through demonstration rather than formal prompt engineering, the app makes it easier to build a personal operational layer for recurring work such as summaries, admin, monitoring, and follow-ups.
CONCLUSION
The screen-recorded skill builder turns everyday desktop routines into repeatable AI actions with minimal setup. Its main appeal is simplicity: users can teach by showing, then automate the task as a reusable command or scheduled job.
Ask a question