An agent input specification describes what an agent accepts as input, declared as InputSpec in the agent module of the Agent Ontology.1 The Agent Ontology declares the class without a definition; this one follows from its position. visionGraph places it under Data Schema and as part of an Agent Operational Profile, which reaches it through input.

It is a declared contract for the data an agent receives, separate from the Agent Trigger that starts the agent. A caller reads it to learn what to send; the agent’s runtime uses it to validate what arrives. The shape of the schema it points to is chosen by the implementer, so it can be JSON Schema, a SHACL shape or any other form.

Its counterpart on the other side of the profile is Agent Output Specification. Both describe the type of data in general, whereas the Input State of an Execution Context records the actual input of one run.

Footnotes

  1. Semantic Agent Communication Community Group, “Agent Ontology” (Agent Ontology, CC BY 4.0). https://github.com/s-agent-comm/agent-ontology/blob/750e06d276dd568313a6c03c4e78203887d3fa4b/ontologies/agent.ttl ↩