Persistence (graph-backed)

Every entry created via create_entry for a type defined here lands as a node in the world graph (the same graph the host’s Cypher / schema-projector / proposition-recall tools talk to). Two consequences worth knowing when writing a realm:

• The entry’s name is the host’s headline for it. The host picks a single field per type to use as the node name for rendering — first title, then name, then summary if any exists. Add at least one of those if you want list / Cypher output to be human-readable.
• The host can add an implicit edge to the acting human principal’s AssistantUser node. The compatibility default is (entry)-[:OWNED_BY]->(principal) — fine for most human-authored types. Service principals have no implicit anchor. When the predicate reads more naturally in the other direction, declare a userAnchor: on the type:

types/movies.yml

- name: MovieRating
  description: "The user's score for a Movie."
  userAnchor:
    predicate: RATED          # uppercase; stored uppercase
    direction: from-user      # `principal -[RATED]-> rating` (default is `from-entry`)
  properties:
    imdbId: "IMDb id of the rated Movie."
    rating:
      type: int
      description: "1–10."

Field	Default	Notes
predicate	OWNED_BY	Cypher-style relationship name. Required when the key is declared. Stored uppercase.
direction	from-entry	from-entry → (entry)-[:PREDICATE]->(principal). from-user → (principal)-[:PREDICATE]->(entry). Pick whichever reads naturally.
(sentinel)	—	Set userAnchor: false to opt the type out of the implicit edge entirely (reference data, type registries, etc.).

Explicit relations between entries

create_entry accepts an optional relations: array so a realm’s skill can wire the new entry to another entry that already exists in the host-bound visibility scope. The host emits each requested edge on save; if any target can’t be found in that scope, the whole create is refused (no orphan node, no orphan edge). A cross-context relation requires an explicit policy-authorized bridge. Example shape, taken from the movie realm’s rate-movie skill:

// inside execute_javascript / execute_python, via the repository tool:
create_entry({
  type: "MovieRating",
  data: { imdbId: "tt0113277", title: "Heat", rating: 9 },
  relations: [
    { predicate: "OF", to: { type: "Movie", imdbId: "tt0113277" } }
  ]
})
// Resulting graph: (User)-[:RATED]->(MovieRating)-[:OF]->(Movie)
//   - The RATED edge comes from `MovieRating.userAnchor` (implicit).
//   - The OF edge comes from `relations:` (explicit, from the skill).

Each relation is {predicate, to: {type, ...keyProps}} — to.type names the target’s type, and the remaining to.* fields are the key properties the host MATCHes against on target nodes in the same host-bound visibility scope. Cross-context matching is never implicit. Use this pattern any time a realm’s typed records form a small connected graph (rating → movie, comment → ticket, note → contact, etc.) — the same graph is then walkable by Cypher and feeds the recall path automatically.