Анализ базы знаний и построение графа

★ 7.1 · knowledge-management

understand-knowledge is a Claude Code skill that analyzes a Karpathy-pattern LLM wiki knowledge base and produces an interactive knowledge graph with entity extraction, implicit relationships, and topic clustering. The expected structure is three layers — raw source documents, wiki markdown files using `[[target]]` wikilink syntax, and a schema file such as CLAUDE.md or AGENTS.md — plus index.md and log.md. Processing runs in five phases: format detection via parse-knowledge-base.py, deterministic scanning into scan-manifest.json, concurrent LLM analysis of 10–15-article batches to surface implicit cross-references, merging with merge-knowledge-graph.py, and saving a validated knowledge-graph.json alongside meta.json. It is built for teams who need to map dependencies across a corporate wiki, discover hidden connections in research documentation, or generate a navigable knowledge map from an existing markdown-based knowledge base.