youtube-to-skill

name	youtube-to-skill
description	Convert any YouTube video into a reusable Claude Code skill. Paste a YouTube URL and this skill fetches the transcript, extracts the actionable content, and writes a ready-to-use .claude/skills/<name>.md file. Best for tutorial videos, how-to guides, workflow demos, or instructional content.
argument-hint	<YouTube URL>

YouTube to Skill

You are converting a YouTube video into a reusable Claude Code skill file. Work through each step in order.

YouTube URL: {{args}}

If {{args}} is empty or contains no parseable YouTube URL, ask the user to provide a YouTube URL before proceeding.

Step 1: Extract Video ID

Parse the video ID from the URL. Handle all common YouTube URL formats:

• https://www.youtube.com/watch?v=VIDEO_ID
• https://youtu.be/VIDEO_ID
• https://www.youtube.com/shorts/VIDEO_ID
• https://youtube.com/live/VIDEO_ID
• https://m.youtube.com/watch?v=VIDEO_ID

Extract VIDEO_ID and confirm it looks valid (11 alphanumeric characters).

Step 2: Fetch Video Metadata and Transcript

Use your web fetch tool to retrieve the pages below (replace VIDEO_ID with the ID from Step 1). Gather both metadata and transcript content:

Source A: YouTube page (metadata: title + description + channel): Fetch https://www.youtube.com/watch?v=VIDEO_ID

Extract from the HTML:

• Video title (<title> tag or "title":"..." in the JSON data)
• Description (from "shortDescription":"..." or the <meta name="description"> tag)
• Channel name
• Duration

Source B: Transcript (primary content source): Fetch https://youtubetranscript.com/?server_vid2=VIDEO_ID

If Source B returns a transcript, use it as the primary content source. If it fails or returns no content, try Source C.

Source C: Fallback YouTube transcript API: If Source B fails, try fetching: https://www.youtube.com/api/timedtext?v=VIDEO_ID&lang=en&fmt=json3

Parse the events[].segs[].utf8 fields to reconstruct the full transcript text.

If both Source B and Source C fail, rely on Source A's title and description only, and note that the transcript was unavailable.

Step 3: Understand the Content

Read through everything collected. Determine:

1. What is this video teaching or demonstrating?
2. Who is the intended audience? (beginner, intermediate, expert; developer, designer, etc.)
3. What is the main workflow or technique?
4. What are the key steps, commands, or instructions the viewer should follow?
5. What tools, technologies, or concepts does it reference?
6. What outcomes does following the video produce?

If the transcript is available, extract all concrete steps, commands, code snippets, and decision points verbatim. Do not paraphrase away specifics.

Step 4: Design the Skill

Before writing, decide:

Skill name: Short, kebab-case, describes what the skill does (not the video title). Example: deploy-to-fly, setup-drizzle-orm, build-realtime-chat.

Skill type: Which pattern fits best?

• Tutorial: Linear steps A→B→C. User follows the video's workflow.
• Reference: Commands, config, or flags from the video organized for quick lookup.
• Workflow: Multi-phase process with decision points and branching.
• Setup: One-time environment or tool configuration.

Does it need {{args}}? Only if the skill is parameterized (e.g., project name, target URL). If the skill is self-contained, omit {{args}}.

Step 5: Write the Skill File

Write the skill to .claude/skills/<name>.md using this structure. The frontmatter at the top must be enclosed in --- delimiters (omit the argument-hint line entirely if the skill takes no arguments):

---
name: <kebab-case-name>
description: <One sentence: what it does and when to use it. Include key tech names so the trigger is precise.>
argument-hint: <optional: what args the user should pass>
---

# <name>: <Human-Readable Title>

<One paragraph: what this skill does, what it produces, and why it exists. Attribute the source: "Based on [Video Title] by [Channel Name].">

**Source:** [Video Title]({{args}}) by [Channel Name]


<Skill body: structured content extracted from the video>

Body content rules:

• Preserve specificity. Keep exact commands, flags, config values, and code snippets from the video. Don't generalize them away.
• Steps should be executable. Each step should be something Claude can actually do or instruct the user to do.
• Use phases for complex workflows. ## Phase 1: X, ## Phase 2: Y
• Use code blocks for every command or code snippet. Even single-line commands.
• Include decision points. If the video says "if you're using PostgreSQL, do X; if SQLite, do Y": keep the branch.
• Strip filler. Don't include the presenter's personal anecdotes, channel intros/outros, or ad reads.
• If transcript was unavailable, note it at the top: > Note: Transcript was unavailable. This skill is based on the video title and description only. Verify steps before running.

Step 6: Confirm and Save

After writing the file, report back:

1. The skill name and file path
2. A one-line description of what the skill does
3. How to invoke it (/<name> or /<name> <args>)
4. Any caveats: e.g., transcript unavailable, video was too abstract to extract concrete steps, commands couldn't be verified

If the video content was too abstract, conceptual-only, or completely inaccessible, tell the user and suggest they paste the transcript manually for better results.