Node reference
Choose the task and understand its binding, data shape and execution limits
Every serialized node has id, kind, label, inputs, outputs and policy, plus its kind-specific section. Resource references select actual workspace/project objects; graph-local node IDs and pool names are not those resource IDs.
| Task | Serialized kind | Section | Execution |
|---|---|---|---|
| Ingress | ingress | ingress | Executable step |
| Transform | transform | transform | Executable step |
| Asset | asset | asset | Special asset path |
| Egress | egress | egress | Executable step |
| Ontology mapping | ontology-mapping | ontologyMapping | Executable step |
| Ontology materializer | ontology_materializer | ontologyMaterializer | Executable step |
| Assertion mapping | assertion-mapping | assertionMapping | Executable step |
| Assertion materializer | assertion_materializer | assertionMaterializer | Executable step |
| Prediction | predict | predict | Executable step |
| MCP tool | mcp-tool | mcpTool | Executable step |
| Query and Skill context | query / skill | query / skill | No default standalone execution step |
Declared interfaces
Ports can declare record, collection, document, blob, ontology-entity, ontology-relationship, ont-fact-set or assertions, plus optional schema/semantic references. Connect actual output and input keys. Ontology mappings normalize to a single out port with ont-fact-set shape.
Configuration examples
Each node page contains Assistant, UI and Graph definition examples. Graph definitions are fragments inside the complete document; there is no standalone public “create this node” HTTP route. Input methods documents the full-pipeline/draft actions instead.
A node policy can declare a write reference, approval metadata, runtime-gate intent, access references and sensitivity. Those declarations do not by themselves establish a per-node human pause. Verify resolved approved policies, actor permissions and the actual execution path.