neo4j-modeling-skill

Start by listing at least five queries the model must answer, then map entities to nodes, connections to directed relationships, and scalars to properties. Apply naming conventions (PascalCase labels, SCREAMING_SNAKE_CASE relationship types) and add uniqueness constraints before any MERGE.

Use a node for things with identity queried as entry points, a relationship for directed connections between two things, and a property for scalars always returned with their parent. If a connection has its own properties or links more than two entities, model it as an intermediate node.

Detect supernodes by querying node degree and flagging nodes with far more relationships than the median for their label. Mitigate by splitting relationship types, segregating labels, applying the bucket pattern for time-series data, or replacing low-cardinality nodes with labels.

No, Cypher query authoring and optimization are explicitly out of scope and handled by the neo4j-cypher skill. This skill covers model design, schema review, constraints, and indexes only; Spring Data Neo4j, GraphQL, and data import are also delegated to separate skills.

Map table rows to nodes, scalar columns to properties, primary keys to uniqueness constraints, and foreign keys to directed relationships. Junction tables with their own columns become intermediate nodes, while simple many-to-many tables become direct relationships.

Storing embedding arrays on business nodes makes them too large and slows traversal queries. Store embeddings on dedicated Chunk nodes linked from the parent document, with a vector index only on the chunk embedding property.