Youtube insights

YouTube Insights

Extract transcripts and deliver structured insight summaries from YouTube videos.

When to Use

• User shares a YouTube URL and wants insights, summary, or transcript
• User says "get me the insight of this video", "resume cette video", "what's in this video"
• User shares multiple YouTube URLs to analyze
• Any request involving YouTube video content extraction

Prerequisites

• yt-dlp must be installed and in PATH (check with where yt-dlp or which yt-dlp)
• Python 3.x available

Workflow

Step 1: Extract Transcripts

For EACH YouTube URL provided, run the extraction script in parallel (one agent per video if multiple):

python "C:\Users\momomuchu\.claude\skills\youtube-insights\scripts\get-transcript.py" "<youtube-url>"

The script returns JSON with: video_id, title, channel, duration, upload_date, view_count, description, transcript.

If the script fails or transcript is empty, try specifying languages:

python "C:\Users\momomuchu\.claude\skills\youtube-insights\scripts\get-transcript.py" "<youtube-url>" "fr,en"

Step 2: Analyze & Summarize

For each video, produce a structured insight summary using this format:

## [Video Title] — [Channel Name]

**Duration:** X min | **Views:** X | **Date:** YYYY-MM-DD

### Key Insights
- [Insight 1 — the most important, actionable takeaway]
- [Insight 2]
- [Insight 3]
- [...]

### What to Take Away
[2-3 sentences: what's personally actionable for the user based on their context/projects]

### Notable Quotes
> "[Direct quote from transcript that captures a key point]"

Rules

1. Parallel extraction — if multiple URLs, launch all transcript downloads simultaneously
2. Language detection — the script auto-tries en, fr, es, de, pt, ja, ko, zh. French videos will get French transcripts
3. Insight focus — don't just summarize chronologically. Extract the NON-OBVIOUS insights, frameworks, numbers, and actionable advice
4. User context — if memory has info about the user's projects/goals, connect insights to what matters to them
5. Be concise — max 10 bullet points per video. Quality over quantity
6. Numbers matter — always extract specific metrics, revenue numbers, conversion rates, timeframes
7. Frameworks — if the speaker shares a step-by-step framework, extract it as a numbered list
8. Warnings — if the speaker warns against something, flag it clearly
9. Skip filler — ignore sponsor segments, self-promotion, "like and subscribe" content