curate-memgraph

name	curate-memgraph
description	Use this to query or change Memgraph by adding canonical entities and relationships, linking Qdrant memories, and keeping graph structure consistent.

Memgraph Curation

Memgraph is the structural source of truth.

Procedure

1. Start from the queries/ files when possible.
2. Use python3 queries/run_query.py <query.cypher> --params-json '{...}' or --params <file.json> for parameterized query helpers.
3. Prefer idempotent graph mutations.
4. Reuse canonical labels and relationships instead of inventing ad hoc structure.
5. Push domain-specific meaning into properties before introducing new graph structure.
6. When a Qdrant memory deserves graph structure, use qdrant/scroll_memories.py or qdrant/get_memory.py to obtain the point ID and normalized packet.
7. Create or update a MemoryEntry node with slug qdrant:<point_id>, then connect it to the relevant entities with the queries/ helpers.
8. After Memgraph updates succeed, annotate the Qdrant point with qdrant/set_graph_link.py.

Canonical Shape

Labels:

• Entity
• Person
• Agent
• Organization
• Place
• Thing
• Concept
• Event
• Work
• Project
• Service
• Collection
• MemoryEntry

MemoryEntry nodes should carry slug, qdrant_id, document_sha256, agent, domain, kind, project, created, summary, linked_at, and optional memory_ref when Qdrant metadata already has one.

Relationships:

• RELATES_TO
• HAS_PART
• INFLUENCES
• LOCATED_IN
• CREATED_BY
• DERIVED_FROM
• OCCURS_IN
• TAGGED_WITH

Standard memory links:

• memory CREATED_BY agent
• project HAS_PART {kind: "memory"} memory
• memory RELATES_TO {kind: "memory-evidence"} selected entities
• replacement memory DERIVED_FROM {kind: "supersedes"} older memory
• memory TAGGED_WITH durable tag concepts

Escalate

• if the requested structure would require a canonical schema change