A video you watch once is knowledge you lose. A Claude skill is knowledge that fires on command, every time, without you rewatching anything. AI Central's Turn Any Video Into a Claude Skill lays out a six step route from a link to a working skill file: pick a video that teaches a repeatable method, mine the transcript with five questions, draft the file with one prompt, split it into two skills, test whether it triggers, then upload it.
Why turn a video into a skill at all
Most people treat a good tutorial video as a one time event. They watch it, nod, maybe take notes, then rebuild the method from memory three weeks later when they finally need it. AI Central's argument is that the method is the valuable part, and the method can be lifted out, written into a form Claude reads, and called by name forever after.
The pipeline is short. A video link produces a transcript, the transcript feeds a single Claude command, and the command produces a skill file. AI Central sums up the payoff in six words.
Build this once, Claude does the rest
That is the entire economic case. The effort is front loaded, once, into an asset that keeps working.
Not every video survives the conversion
AI Central is blunt about the input. Skills should come from videos that teach a clear method, system, framework, checklist, tutorial or repeatable process. Vlogs, opinion pieces, reaction videos and entertainment clips are ruled out, along with anything that has no clear steps.
The filter AI Central offers is one question you can answer before investing any time. Can you summarise the video's method in numbered steps? If yes, it converts. If no, it will not, and the reason is structural rather than a matter of taste. A skill is a set of instructions. A video with no ordered steps has no instructions to hand over.
This is the step most people are tempted to skip, and it is the one that decides whether everything downstream was worth doing.
Watch with five questions in hand
The second step is pulling the transcript, and AI Central attaches five questions to ask while the video is still playing.
- What problem does this video solve?
- What are the steps shown in order?
- What rules does the creator keep repeating?
- What mistakes does the creator warn people about?
- What exactly should Claude produce after using this method?
Read as a set, those five are not a note taking template. They map onto the anatomy of a working skill file. The first is the trigger condition. The second is the procedure. The third becomes the constraints the model has to hold to. The fourth becomes the failure modes, the negative instructions that stop Claude drifting back into generic advice. The fifth is the output contract, and it is the one people forget.
AI Central words that last question in the most demanding form available.
What exactly should Claude produce after using this method?
The word exactly is doing real work there. A skill that knows the steps but not the deliverable produces a lecture instead of a result.
One prompt does the drafting
The third step hands the job to Claude. AI Central supplies a single copy and paste prompt that calls Claude's skill creator, gives it the video title, the creator's name, the link and the full transcript, then names what to extract: the main problem the video solves, the step by step method, the rules the creator repeats, the mistakes to avoid, and the output format Claude should follow.
Two lines in that prompt carry more weight than the rest. The first is a negative scope list. AI Central tells Claude the skill is not for general advice, not for random summaries, and not for tasks unrelated to the video's method. Skills fail far more often by over firing than by under performing, and that short list is the cheapest defence available.
The second tells Claude to interview you on anything missing before it generates the file, and to run an evaluation afterwards. That turns a one shot generation into a short interrogation, so the gaps a rough transcript leaves get filled in by you rather than invented by the model.
Two skills beat one
The fourth step is AI Central's sharpest structural call, and it is the easiest to miss.
Long videos often contain two separate things.
The recommendation is to split them. One skill diagnoses, working out what is wrong using the video's framework. The other applies, fixing or rewriting using the video's exact method. AI Central names the pair after the video, one ending in diagnose and one ending in apply, and the worked example it gives comes from a video about writing better hooks.
The logic holds well beyond that example. Judgement and execution are different jobs with different inputs. A diagnostic skill needs the framework and a critical eye. An application skill needs the method and a target output. Fuse them into one file and you get a model that critiques when you wanted a rewrite, or rewrites when you wanted a verdict.
Test the trigger, not the output
The fifth step is the one almost everyone skips, and AI Central reduces it to three checks. Ask for the same thing five different ways and confirm the skill fires every time. Make three unrelated requests and confirm it stays quiet. Then ask Claude itself when it would reach for the skill.
Notice what is under test. Not the quality of the output, the routing. A skill that never fires is invisible, and a skill that fires on everything is worse than no skill at all, because it drags an irrelevant framework into unrelated work.
The third check is the cleverest of the three.
Ask Claude directly: "When would you use this skill?"
Asking the model to describe its own trigger conditions shows you how it actually read your description, which is rarely how you thought you wrote it. If the answer comes back vague, the description is vague, and the fix belongs in the file rather than in your next prompt.
Installing it takes four clicks
The last step is mechanical. In Claude, open Settings, then Capabilities, then Skills, then Upload. From that point AI Central says the skill is available by slash command anywhere in Claude, and it runs automatically once called.
That is the moment the asset starts paying back. The video is no longer something you have to remember. It is something you invoke.
Where to start
If you have never built one, take a single video whose method you already trust and run all six steps end to end without optimising anything. The point of a first pass is to watch the loop close, not to produce a masterpiece.
If you already have skills installed, the highest leverage change here is the split. Go back to your longest, vaguest skill and ask whether it is quietly doing two jobs. If it is, break it into a diagnose skill and an apply skill, then retest both against AI Central's three checks.
And if a skill you built refuses to fire, do not rewrite the method. Rewrite the activation description and run the five ways test again. The method is almost never the problem.
Do I need a special tool to get the transcript?
AI Central's process starts from the video's transcript and its link, and it does not prescribe a particular extraction tool. What it does prescribe is what you are hunting for while the video plays: the problem, the ordered steps, the repeated rules, the warned against mistakes, and the exact output Claude should produce.
What kind of video actually works?
Anything that teaches a clear method, system, framework, checklist, tutorial or repeatable process. AI Central rules out vlogs, opinion pieces, reaction videos and entertainment clips. The test is whether you can write the method out as numbered steps.
How do I know if my skill actually works?
Test the trigger before you judge the output. Ask for the same thing five different ways to see whether it fires each time, make three unrelated requests to see whether it stays silent, then ask Claude when it would use the skill.
Should every video become two skills?
No, only the long ones carrying two jobs. AI Central's split applies when a video both diagnoses a problem and prescribes a fix. A short video teaching one method is fine as a single skill.
Where does the finished file go?
Into Claude through Settings, then Capabilities, then Skills, then Upload. After that you call it with its slash command anywhere in Claude and it runs on its own.