Drop a YouTube URL, get a Reel script or a 7-slide carousel in your brand voice. This guide gets you set up in about five minutes.
Download agent youtube-to-social.skill · 1.7 MBOne file: youtube-to-social.skill. It's a zip archive with a special extension that Cowork recognizes as installable. Inside are the brain (SKILL.md), the format library (29 viral video formats), the carousel composer, six bundled fonts, and two helper scripts.
You don't need design software, fonts, or any account. Once installed, you talk to it in plain English.
Claude with Cowork mode, or Claude Code CLI on your computer.
An internet connection. The skill fetches YouTube transcripts on demand.
An image-generation MCP if you want AI illustrations on carousels. Higgsfield works best.
No API keys, no design tools, no font installs. The skill brings everything it needs.
Worth knowing: Without an image-gen MCP, the carousel falls back to a clean text-only design that always works. You won't get blocked, you'll just have a different aesthetic.
Three ways. Pick whichever feels easier.
Find youtube-to-social.skill in your Downloads folder.
The file appears as a card with a Save skill button.
Cowork installs it into your skills directory and confirms.
Click Customize in Cowork to open your skills and plugin settings.
Pick youtube-to-social.skill from your file picker.
Cowork installs the skill and lists it in your library.
Same result, different path. Drag-and-drop is faster for one-off installs. The Customize menu is handy if you want to manage all your skills in one place.
If you're a developer using the Claude Code CLI instead of Cowork, the install is manual but quick.
Change youtube-to-social.skill to youtube-to-social.zip if your OS won't unzip .skill directly.
You'll get a folder called youtube-to-social/ with SKILL.md inside.
For a global install: ~/.claude/skills/youtube-to-social/. For a per-project install: your-project/.claude/skills/youtube-to-social/.
The skill loads automatically and is ready to use.
The first time you use the skill, it asks four quick questions. Your answers save to a config file inside the skill folder, and you won't be asked again.
1. Brand name. What your brand is called. Example: "Honest Lighting".
2. Audience. One sentence on who you make content for. Example: "YouTube creators learning to look pro on a budget".
3. Voice. Pick one of: Punchy & casual, Polished & professional, Warm & conversational, Technical & precise, or describe your own.
4. Default closing CTA (optional). Example: "Save this", "Follow for more", "Comment your take". Leave blank if you don't want one.
That's it for basic setup. The first time you make a carousel, you'll be asked four more visual questions (aspect ratio, font, visual mode, aesthetic). Also asked once.
Want to change something later? Just say "redo setup" or "change my voice" or "redo carousel settings" in chat. The config is editable any time.
Paste a YouTube URL into chat and tell the skill what you want. Examples:
make a reel from https://www.youtube.com/watch?v=...carousel from https://youtu.be/...both from https://www.youtube.com/watch?v=...
Or paste the URL on its own and the skill asks which output you want.
The skill fetches the transcript, mines it for the strongest substance (claims, frameworks, specific examples, common mistakes), and asks you to choose an angle.
For scripts, you get 2 to 3 viral format options. Each one comes with a one-line explanation of how it would apply to this specific video. Pick the one that feels right, and the skill writes a 30 to 60 second talking-head script with on-screen text and B-roll cues.
For carousels, the skill derives the shape from the substance (a tier walk, a list of mistakes, a story arc) and writes 5 to 7 slides. You get individual PNGs for Instagram, a stitched PDF for LinkedIn, and a single grid preview.
The output is yours, not a recap. The skill never references the source video. No "in this video" or creator credit. It uses the transcript as raw material, then writes original content in your voice.
Carousels can run in two visual modes:
| Mode | What it looks like | What it needs |
|---|---|---|
| Text-only | Clean designed slides on a colored gradient. Premium typographic look. | Nothing. Always works. |
| AI image + text overlay | A custom-generated illustration or photo per slide, with text overlaid in post. | An image-gen MCP connected to Claude. |
Higgsfield is the recommended provider. Their MCP gives you photorealistic, illustrative, and stylized image generation, and it's built for marketing-grade output. Setup takes about two minutes.
Connect Higgsfield: Follow the official MCP setup guide at https://higgsfield.ai/mcp. Once connected, the skill detects it automatically and uses it for any carousel run set to image-overlay mode.
If your image-gen MCP fails on a specific run (rate limits, content filter, model hiccup), tell the skill "text-only carousel" and it falls back to the always-works design for that one run.
| Say this | What happens |
|---|---|
redo setup | Re-asks the four basic onboarding questions. |
change my voice | Updates just the voice preset. |
redo carousel settings | Re-asks aspect ratio, font, visual mode, aesthetic. |
you pick | The skill picks the sharpest-hook format for you instead of asking. |
make a [format name] from this | Skips the menu and uses the format you named. |
text-only carousel | Forces no-AI-image mode for this run. |
use [aesthetic] for this one | Per-run aesthetic override without changing your default. |
Drop two or more YouTube URLs in a single message and the skill enters bulk mode. It auto-picks formats, processes everything in parallel, and delivers a batch with a summary table at the end. No per-video questions. Use this when you've got a list of videos to turn into content fast.
Captions disabled on a video. Some YouTube videos lock transcripts. Paste the transcript text into chat and the skill picks up where it left off.
Image generation failed. Either no image-gen MCP is connected, your provider rate-limited, or a content filter triggered. Say "text-only carousel" to fall back, or "regenerate that slide" with a slightly different prompt.
Wrong voice or audience saved. Say "redo setup" or "change my voice." The config is editable any time.
Skill doesn't appear in Cowork. Confirm the file ends in .skill (not .zip). If you renamed it during install, rename it back. Restart Cowork and try the install again.
Your config file (brand, audience, voice, CTA) stays on your machine inside the skill folder. The skill makes outbound calls only to YouTube to fetch a transcript and to your chosen image-gen MCP if you've connected one. No analytics, no telemetry, no tracking.