Hybrid search combining vector similarity with graph traversal for multi-hop reasoning over the knowledge graph
Hybrid search combining vector similarity with graph traversal for multi-hop reasoning over the knowledge graph
npx skills add https://github.com/okikusan-public/knowledge_graph --skill graph-search复制此命令并粘贴到 Claude Code 中以安装该技能
Autonomous multi-tool search agent that dynamically selects search strategies, evaluates results, and iterates until sufficient context is gathered to answer complex questions
Search X (Twitter) posts via Grok API and ingest results into the knowledge graph.
Convert YouTube video to structured Markdown via markitdown (metadata + transcript) and ingest into the knowledge graph.
Add knowledge directly to the graph from free-form text input without requiring a document file. Supports Note, WebSource, Conversation, or any custom source type.
Convert PDF to structured Markdown via markitdown and ingest into the knowledge graph. Preserves tables, headings, and formatting better than plain-text extraction.
Spaced repetition quiz using GraphRAG entities. Generates questions, grades answers, and tracks learning progress with optimal review intervals.
| name | graph-search |
| description | Hybrid search combining vector similarity with graph traversal for multi-hop reasoning over the knowledge graph |
Hybrid search that combines vector similarity with graph traversal to discover and synthesize knowledge across the full graph structure.
/graph-search "What technologies are used in the project?"
/graph-search "generative AI projects" --project project_a
/vector-search: Returns top-K similar nodes (isolated matches)/graph-search: Finds seed nodes via vector search, then traverses RELATES_TO, BELONGS_TO, MENTIONS, and SOURCED_FROM to gather multi-hop context, then synthesizes a coherent answerpython ${CLAUDE_SKILL_DIR}/../../../scripts/graph_search.py "$QUERY" $1 $2 --json
This returns JSON with:
seed_entities: Top entities by vector similarityseed_chunks: Top chunks by vector similarityseed_communities: Top communities by vector similarityrelated_entities: Entities connected via RELATES_TO (graph expansion)entity_communities: Communities the entities belong tocontext_chunks: Source text chunks mentioning relevant entitiesprovenance: Which documents contributed each entityIf the context chunks from Step 1 are truncated or insufficient, read the full text of the most relevant chunks. Use the source_path from provenance to read original documents if needed.
Using all gathered context (seed nodes, related entities, communities, chunks, provenance), synthesize a comprehensive answer to the user's question:
At the end, list the source documents and key entities that informed the answer.
--top / -k: Number of seed nodes per type (default: 3)--max-related: Max related entities to expand (default: 10)--max-chunks: Max context chunks (default: 10)--project / -p: Project name