What problem does it solve? Traditional RAG rediscovers knowledge from scratch on every query, losing cross-references and synthesis work. This Skill builds a persistent Karpathy-style LLM Wiki — a directory of interlinked markdown files that compiles knowledge once and keeps it current, so contradictions are flagged and synthesis accumulates over time. ## Core Features & Use Cases - Source Ingestion: Capture URLs, PDFs, and pasted text into an immutable raw/ layer with sha256 drift detection, then synthesize cross-referenced entity and concept pages. - Domain Querying: Answer questions by reading the compiled wiki, citing pages, and filing valuable syntheses back into queries/ or comparisons/. - Wiki Linting: Audit for orphan pages, broken wikilinks, stale content, contradictions, tag taxonomy drift, and source integrity. - Use Case: A researcher tracking AI/ML papers ingests each new arxiv paper; the agent updates 5-15 interlinked pages, maintains the index and log, and answers future questions from the compiled knowledge instead of re-reading every source. ## Quick Start Ask the agent to initialize a new wiki for your research domain and ingest your first source URL into it.