Turn YouTube videos into Claude skills

Here's how to set up what I showed in my video. It takes two steps, and you won't write any code.

What it does

Claude can't watch a video on its own. A free skill called watch gives it a way: it pulls still frames and the transcript out of a YouTube video, so Claude can see and read what's in it.

Then you ask Claude to turn what the video teaches into a skill of its own. A skill is a saved set of instructions that Claude can use again whenever you need it. So you don't just get a summary you'll forget. You get something you can reuse.

What you need

• A paid Claude plan. Claude Code comes with Pro, Max and Ultra (and with Team and Enterprise). It isn't in the free plan.
• The Claude desktop app. The skill runs in Claude Code, which is the Code tab of the app, not the regular chat.
• A computer. I set this up on a Mac. The skill says it supports Windows too, but I haven't tried it there.

Using Codex instead? It works there too, and the steps are at the end of this guide.

Before you start: open Claude Code

1. Download the Claude desktop app from claude.ai/download, install it and sign in.
2. Click Code at the top of the sidebar.
3. Make a new folder to work in, for example "Claude skills" in your Documents.
4. In the row under the message box, choose Local, then click the folder control, choose Open folder… and pick that folder.
5. In the same row, choose Auto, so Claude can keep going without stopping to ask at every step.

Keep an eye on it. In Auto, Claude can install programs and edit or change files without asking first. If you'd rather approve each step, choose Manual.

Step 1: install the skill

Paste this into the message box and press Enter:

Install the watch skill from https://github.com/bradautomates/claude-video for me, then check that everything it needs is on my computer.

Claude adds the skill for you. The skill also needs three free programs on your computer: Python, ffmpeg and yt-dlp. Claude checks for them and can install the ones that are missing.

On a Mac, the skill installs them with Homebrew, a free installer for this kind of program. If your Mac doesn't have Homebrew yet, that one step is yours: Homebrew's installer asks for your Mac password, which Claude can't type for you. Claude can tell you what to paste into the Terminal app.

When it's done, start a new session in the Code tab so the skill loads.

Rather click through it yourself? The skill's GitHub page shows how to install it with clicks, under "Claude Desktop".

Step 2: watch a video and turn it into a skill

1. Type /watch and pick what comes up, usually /watch:watch.
2. Paste a YouTube link after it, and say what you want. For example:

/watch:watch [your YouTube link] Watch this, learn what it teaches, and turn it into a skill I can use forever.

Here's how I'd ask with Seth Godin's interview on The Diary of a CEO, the one from my video:

/watch:watch https://www.youtube.com/watch?v=g7AxxkywiFI Watch this, learn what it teaches about getting unstuck, and turn it into a skill that helps me when I'm overthinking.

The first time, it asks how to watch videos.

• Local needs no account and no key. The frames and the transcript are made on your computer, and nothing is sent to Google. The transcript comes from the video's captions, which most YouTube videos have. The skill marks Gemini as recommended, but this is the one I use.
• Gemini lets Google's AI watch the whole video, sound included. It needs a free key from Google AI Studio, and your videos are sent to Google.

If you choose Local, it asks two more things. For detail, pick balanced. For videos without captions, it recommends whisperx, a 1.5 GB download. Pick none to start. You can add a way to transcribe those later.

Then make the skill yours.

• Read the skill Claude writes, and ask for changes until it fits how you work.
• Ask Claude to save it as a personal skill, so it works in any folder you open.
• Next time, ask for it by name, or type / and its name.
• For a podcast or an interview, where the picture is mostly people talking, ask for the transcript only. It skips the frames.

Using Codex instead of Claude Code

The watch skill has its own Codex install. You'll need a ChatGPT account and the ChatGPT desktop app. These steps come from the skill's GitHub page and OpenAI's docs. I set this up in Claude Code and haven't run it in Codex myself.

1. Download the ChatGPT desktop app from chatgpt.com/download, install it and sign in.

2. Choose Codex from the ChatGPT dropdown, then open a new folder to work in.

3. In the permissions control below the message box, choose Approve for me, so it can keep going without stopping to ask. Keep an eye on it, because it can install programs and change files.

4. Paste this into the message box and press Enter:

   Use $skill-installer to install the watch skill from:
   https://github.com/bradautomates/claude-video/tree/main/skills/watch
   

5. Start a new chat so the skill loads. If it doesn't show up, restart the app.

Then follow Step 2. Instead of typing /watch:watch, ask for the skill by name. For example:

Use the watch skill on [your YouTube link]. Watch this, learn what it teaches, and turn it into a skill I can use forever.

Checked on 6 November 2026 with version 0.3.2 of the watch skill, on a Mac. Claude changes fast: if a menu has moved, look for the same words in the app.