Have an agent watch a tutorial, list the steps, and save a skill only after you review

The scene

You have a half-hour how-to video, and what you actually want is a repeatable procedure rather than another viewing. You want a coding agent to list the steps, commands, and likely failure points first, and to save a skill only after you have checked that list.

Tools involved

What they did

1. Use only a tutorial you have the right to use. Do not hand over paid courses or videos whose terms forbid saving or extracting them.
2. Give the video link to a coding agent. If you use a tool that streams picture and audio locally, confirm the media stays on your machine instead of being uploaded somewhere by default.
3. Ask first for a list: each step, the commands or screens it uses, how you will know it worked, and any mistakes the presenter ran into. Do not install a skill or change your project yet.
4. Check that list against the video or your own goal. Correct misread screens, missing steps, and outdated commands.
5. Only then have it turn the list into a reusable skill or note, and try the skill on a small real task. If the trial fails, fix the skill before you use it on anything that matters.

Copyable prompt

Watch the tutorial at [link I have the right to use]. Do not change any of my files yet. Extract: (1) the goal of the tutorial, (2) a numbered list of steps, (3) each command or screen the presenter uses, (4) how to tell a step succeeded, and (5) mistakes or workarounds shown in the video. If the picture and the narration disagree, say so and do not guess. Stop after the list so I can review it. After I approve, save it as a skill named [name] and wait; I will pick a small task to try it on.

Rewritten here for you to try; not a verbatim excerpt from elsewhere. What it asks：Extract a checkable step list first and change no files. After you approve, it may save a skill, and you choose a small trial task.

Things to watch

The author said the open-source tool streams picture and audio locally and does not send large files to an outside API; that was not retested here, and it may not work on every video site. An agent can misread a button, miss a spoken limit, or treat an outdated screen as current. Do not extract paid or restricted videos. A person should review the list before it becomes a skill. Generally reproducible in China if a coding agent can open the video.

You need a coding agent. Locally streaming a video also uses disk and time, depending on the length of the video.