Media Extract

What it does

Media Extract is a utility skill that ingests content from YouTube, web pages, local files (video, audio, PDF, text) or pasted transcripts and returns structured, parseable outputs: summaries, chapter timestamps, golden nuggets, quotes, command/code examples, and visual analysis via Gemini. It can download videos using yt-dlp, clean meeting transcripts (remove timestamps and filler words), and enrich outputs with metadata (channel, publish date, duration, views, likes, engagement). The skill focuses on consistent, machine-friendly formats so downstream skills can consume the results programmatically.

When to use it

Use this skill when a user shares a YouTube link, posts an article URL, uploads or points to a local media file, pastes a transcript, or asks explicitly to "clean this transcript", "remove timestamps", or to "analyze this video/article". Also useful for batch processing playlists or folders of videos and for extracting code shown visually on-screen.

What's included

• Scripts: instructions reference yt-dlp-based fetch scripts and a Python extract script for Gemini visual analysis (requires google-genai).
• References: workflow files for summary, visual-analysis, chapters, cleaning transcripts, and batch processing are outlined.
• Instructions: source-detection rules, metadata enrichment, caching behavior for transcripts, and output naming conventions are provided in the skill body.

Compatible agents

Best suited for agents with file and web access and Gemini integration (e.g., Claude with file+tooling, agents using Gemini visual models, or other assistant runtimes that can run local scripts and call external APIs).