What problem does it solve? Researching a topic across dozens or hundreds of YouTube videos normally means watching or reading everything, which floods context with millions of characters. This Skill collects candidate videos, fetches transcripts in parallel, verifies upload recency, and pattern-matches across the corpus so only the most relevant videos get deep-read. ## Core Features & Use Cases - Parallel transcript extraction: Uses yt-dlp and the instance-based youtube-transcript-api with ThreadPoolExecutor (12-16 workers) to fetch hundreds of transcripts quickly. - Recency verification: Enforces a maximum age in days by fetching upload dates with yt-dlp, catching gaps between YouTube's search filters and actual upload dates. - Pattern-based synthesis: Splits transcripts into sentences, scans for keyword patterns, ranks videos by hit count, and deep-reads only top-cited sources instead of dumping everything into context. - Use Case: Ask "what's new about AI agents this week" and get a synthesized report organized by topic with citations, drawn from 200+ videos all confirmed under 7 days old. ## Quick Start Use the bulk-youtube-research skill to find everything new about a topic on YouTube from the past week and synthesize the findings with citations.
