This implements Andrej Karpathy's llm-wiki methodology: compile knowledge once from articles, PDFs, YouTube videos, and WeChat posts, then maintain it as a structured wiki instead of re-processing raw documents on every query. It auto-routes sources to the right extraction tool (baoyu-url-to-markdown for web content, youtube-transcript for videos), generates Obsidian-compatible pages with bidirectional links, and runs health checks for orphaned pages and contradictions. Works across Claude Code, Codex, and OpenClaw. If you're drowning in bookmarks and PDFs but keep re-reading the same content because you can't remember what mattered, this gives you a actual persistent knowledge base that improves over time.
npx skills add https://github.com/aradotso/trending-skills --skill llm-wiki-skill