/kg

Knowledge graph commands:

kg extract <path> -- Extract entities and relations from source files at the given path.

1. Scan files at <path> recursively for classes, functions, modules, types, and config references
2. For each entity, record its type, name, file location, and description
3. Map relations between entities: imports, extends, implements, depends-on, calls, references
4. Store entities via mcp__claude-flow__agentdb_hierarchical-store in the knowledge-graph namespace
5. Create causal edges via mcp__claude-flow__agentdb_causal-edge for each relation
6. Report: total entities found, total relations mapped, entity type breakdown

kg traverse <entity> -- Pathfinder traversal starting from the named entity.

1. Look up the seed entity via mcp__claude-flow__agentdb_hierarchical-recall
2. Expand outward by following causal edges (default depth: 3)
3. Score each path: relevance = edge_weight * semantic_similarity(query, node)
4. Prune paths with cumulative score below 0.3
5. Return the top 10 paths with entities, relations, and scores

kg relations <entity> -- List all direct relations for the named entity.

1. Query causal edges where source or target matches <entity>
2. Group by relation type (imports, extends, implements, depends-on, calls)
3. Display as a table with: relation, direction (incoming/outgoing), target entity, weight

kg visualize -- Generate an ASCII visualization of the knowledge graph.

1. Recall all entities and edges from the knowledge-graph namespace
2. Identify the most-connected nodes (top 10 by degree)
3. Render a simplified graph showing key nodes and their connections
4. Include legend with entity types and relation types

kg search <query> -- Semantic search across the knowledge graph.

1. Search entities via mcp__claude-flow__agentdb_pattern-search (the semanticRouter controller is enabled: false in current AgentDB builds — pattern-search is the available substitute; see ruvnet/ruflo#2049)
2. Expand results with causal edges to show related context
3. Rank by pattern-match score (fall back to mcp__claude-flow__embeddings_generate + manual cosine for higher-fidelity semantic similarity if needed)
4. Display matches with entity name, type, file location, and relevance score