SKILL -- graphify

Trigger conditions

Use this skill when the conversation involves:

• Storing knowledge, concepts, decisions, or entities in the graph
• Retrieving related concepts via graph traversal or semantic search
• Querying "what is connected to X", "what decisions relate to Y"
• Indexing research output from deep-research-core
• Seeding the graph from Obsidian vault notes
• Running a LangGraph research -> extract -> store -> reflect loop

What graphify does

graphify is a hybrid knowledge graph combining Neo4j (structural) and Qdrant (semantic).

Write path:

1. Receive content (text, metadata, node type)
2. Create/update Neo4j node with relationships
3. Embed content -> Qdrant vector (same ID as Neo4j node)

Read path:

1. Query arrives (natural language or structured)
2. Parallel: Neo4j graph traversal + Qdrant vector search
3. Merge + rank results
4. Return enriched context to caller

Node types

Concept | Entity | Task | Skill | Agent | Source | Session | Decision

MCP tools (when graph-server is active)

• graphify_store(node_type, title, content, relationships[])
• graphify_retrieve(query, mode="hybrid", limit=10)
• graphify_relate(from_id, rel_type, to_id)
• graphify_ping()

LangGraph state mapping

Concept -> state["knowledge"] Task -> state["tasks"] Decision -> state["context"] Source -> state["sources"]

Dependencies

• Neo4j port 7687 : docker compose up -d neo4j
• Qdrant port 6333 : docker compose up -d qdrant
• NEO4J_URI, NEO4J_USER, NEO4J_PASSWORD, QDRANT_URL in .env

Status

v0.1 scaffold -- planning only. Implementation in Phase 3. Use validate_repo.py --show-stubs to track progress.