extract-docslisted

# Extract Documentation Knowledge (Sonnet) **Purpose**: Systematically extract knowledge from scattered documentation to prevent catastrophic forgetting. ## When to Invoke Autonomously 1. **Project has 20+ scattered documentation files** - Risk of losing context 2. **Before archiving completed work** - Extract still-relevant content first 3. **Preparing canonical living documents** - Consolidate before creating 4. **After major phase completion** - Capture learnings before forgetting ## DO NOT invoke for - Well-organized documentation (no consolidation needed) - Active editing sessions (extract afterward) - Single-file documentation tasks ## Search Systems **1. kg-search/kg-info** (Keyword/Metadata) - Fast (~100ms): - `.claude/scripts/kg-search search "term" [--type TYPE] [--tags TAGS]` - `.claude/scripts/kg-info info "Title"` **2. Weaviate MCP Tools** (Semantic/Graph): - `search_knowledge_graph` - Basic semantic (~500ms) - `semantic_graph_search` - GraphRAG with WikiLink traversal (~1-2s) - `hybrid_search` - Parallel keyword+semantic+graph (~1-2s) **3. Code Graph** (Semantic Code Search): - `search_code_graph` - Find code by purpose/concept (~200-500ms) - `query_code_structure` - Dependencies, callers, inheritance (~50-100ms) - CLI: `.claude/scripts/code-graph-query search "pattern"` **Decision**: Known terms → kg-search | Concepts → search_knowledge_graph | Relationships → semantic_graph_search | Code entities → search_code_graph ## What This Skill Does ### Docu