learn-skill

Video and podcast content is everywhere, but turning a forty-minute talk into a structured, searchable note still means manual transcription, screenshot OCR, and reformatting. learn-skill automates the entire pipeline as an agent-native workflow: hand it a shared link or a local audio file, and it normalizes short URLs, extracts subtitles (falling back to resilient transcription when none exist), runs keyframe OCR, then performs a three-stage Map-Reduce-Verify pass using the current agent session model. No external model API calls are made; the main script only handles local extraction and evidence storage.
The output is a hierarchical knowledge card rather than a wall of text: timestamped chapter breakdowns, a Mermaid mind-map, highlighted insights, glossary terms, flashcards, and a three-part summary. Import straight into Obsidian (default) or SiYuan, or keep a local Markdown copy. A SQLite task ledger tracks every artifact for resumable runs, and local scratch files are cleaned up after a successful import so your vault stays tidy.
Built for developers and researchers who already work with AI agents and maintain a personal knowledge base. No extra model credentials, no tool-chain juggling: paste a link, give the agent a one-line instruction, and a well-structured, reviewable note lands in your vault minutes later.