What problem does it solve? Chat-based research disappears when the session ends, forcing you to re-synthesize the same sources repeatedly. This Skill implements Karpathy's llm-wiki pattern: a persistent, compounding knowledge base where raw sources stay immutable and the LLM owns a cross-linked markdown wiki that accumulates across sessions. ## Core Features & Use Cases - Three-Layer Architecture: Separates immutable raw sources, LLM-owned wiki pages (entities, concepts, sources, queries), and a CLAUDE.md schema file defining conventions. - Ingest / Query / Lint Operations: Stage sources into raw/ and generate summary pages, answer questions with citations and file valuable answers back into the wiki, and run health checks for orphan pages, broken cross-links, and stale content. - Index and Log Conventions: Maintains a content-oriented index.md catalog and an append-only, grep-parseable log.md journal of every wiki activity. - Use Case: A researcher ingests articles and papers over weeks; each ingest updates related entity and concept pages, so a later question like "RAG vs llm-wiki differences" is answered from already-synthesized, cross-referenced pages instead of re-reading raw sources. ## Quick Start Ask the assistant to initialize your personal knowledge wiki and ingest your first article or document into it.
