Most people treat “learning a skill” and “using a tool” as two separate jobs, and both land on them. There is a third option: make the tool do the learning. Claude cannot watch YouTube out of the box, but that is a gap you close once, and after that any video with a technique in it becomes something Claude can absorb and keep.
The shift sounds small and changes your week. Watching a good video used to produce a feeling of having learned something, which then evaporated because you never used it. When Claude learns instead, the video produces an artifact: a skill, stored, named, and available the next time the situation comes up. The emotion is gone and the asset remains, which is the correct trade.
Teach Claude to watch, once
The capability is a skill, and installing it is a single instruction. Point Claude at the skill source and tell it to install the skill that teaches it how to watch videos and pull accurate transcripts. That is the whole setup. It takes seconds, and you only do it one time. Everything after this is the same three-line request applied to a different video.
Then any video becomes a lesson
Once the skill is in place, drop in any YouTube video and give the instruction that matters: watch this video, learn the core elements of what it is teaching, and turn that into a skill I can use in the future. Video editing, sales outreach tactics, travel planning, it does not matter. The request is the same, and the video is the input. The consistency is the point: you have turned an open-ended task into a repeatable one.
Ask it to also tell you the three situations the skill applies to. That last question is a cheap verification step. If Claude cannot name when the skill would be used, it has summarised the video rather than learned it, and a summary is not a skill. The follow-up catches the difference before you lean on it.
A skill library that builds itself
Handled this way, Claude both learns the technique and turns the learning into a reusable skill it can apply later, without the original video. Every good video you find compounds into your own library, and the library is the asset. You are no longer the bottleneck between watching and doing, because you are not the one doing the learning. Over a month of ordinary scrolling, you can accumulate a working reference of techniques that would have taken a course to assemble.
What to watch
Two honest limits. A transcript can be wrong, so a skill built from one bad video will carry the error; give Claude a second source when the topic matters. And a reusable skill is only as good as its trigger, so name it clearly and describe when to use it, or it will sit unread while you do the work by hand.
There is a third thing worth saying plainly: not every video deserves this. A technique worth keeping is one you will plausibly need again. Turning a one-off curiosity into a skill just adds clutter to a library you have to search. Be a little ruthless about what earns a place.
The shift is small and it changes your week: stop being the one who learns, and make the learning a thing your system keeps.
How the library compounds
A single skill is useful. A library of them changes the shape of the work, because the second skill is cheaper than the first. Once Claude knows how you like a task done, the next skill in that area inherits the vocabulary, the tone and the constraints, and you stop re-explaining the fundamentals every time. The tenth video you feed it teaches more than the first, because it lands on top of context that already exists.
The practical habit is to review the library occasionally and delete what you never used. A skill you have not invoked in a month is usually a curiosity you mistook for a capability, and every one of those makes the useful ones harder to find. Activate the right skill at the right time and the system stays sharp, which means the system stays untended.
If you do one thing after reading this, do the smallest possible version: install the watching skill, drop in a single video you have been meaning to watch, and turn it into one skill. The mechanism is unremarkable. The compounding is not, and you only see it after the first one exists.