YouTube Video Skills

This workspace bundles two CLAUDE style skills for discovering YouTube content and, when needed, capturing transcripts.

Use cases

• AI‑assisted research from YouTube
  Search for talks, panels, or lectures on a topic and extract clean English transcripts for downstream analysis, summarization, or citation.

• Topic discovery & video curation
  Rank and group relevant YouTube videos by inferred subtopic, producing lightweight markdown reports without downloading transcripts.

• LLM / agent workflows
  Provide structured, repeatable inputs (queries, metadata, transcripts) for Claude, Codex, or similar agents in RAG, summarization, or report‑generation pipelines.

• Knowledge base creation
  Build organized, topic‑based corpora from conference talks, interviews, and long‑form videos for internal knowledge management or search.

• Event and landscape monitoring
  Track emerging themes and speakers across conferences or communities by periodically searching, grouping, and capturing video content.

Skills

• skills/youtube-search-transcript/SKILL.md — Complete workflow that searches YouTube, groups results by topic, creates topic folders under outputs/, and captures transcripts/captions for downstream work.
• skills/youtube-topic-search/SKILL.md — Lightweight workflow that focuses on ranking and reporting relevant videos for a topic without downloading transcripts.

Quick Start

1. Create a .env file with YOUTUBE_API_KEY=<your_api_key>.
2. Install tooling: yt-dlp, python-dotenv, youtube-transcript-api, google-api-python-client.
3. Ensure an outputs/ directory exists. The transcript skill will add topic subfolders (e.g., outputs/claude-code/); the search-only skill writes markdown reports here.
4. Trigger the skill that matches your task:
   • youtube-search-transcript when you need transcripts and organized topic folders.
   • youtube-topic-search when you only need curated video recommendations.

Follow each SKILL.md deliverables checklist to capture queries, metadata, and outputs consistently.

Example

Use <install_path>/youtube-search-transcript to search for "AI safety conference talks".

Group the videos into a small set of inferred subtopics (max 6). For each subtopic:

• Save cleaned English transcripts to outputs//<video_title>.txt
• Ensure transcripts are well-formatted and readable

Then, for each subtopic, produce a summary table with:

• Title
• Speaker(s)
• 3–5 key points
• Video URL

Rules:

• Skip videos without English transcripts
• Do not duplicate videos
• Assign each video to only one subtopic

If the tool is located in ./Claude/skills, call: /youtube-search-transcript