Define what an autonomous agent is — structurally.

Agents without structure cannot be governed

Every agent framework defines an agent implicitly — as whatever the framework instantiates when you call its constructor. There is no canonical definition of what fields an agent must have, what types those fields must be, or what structural invariants must hold for an agent to be considered valid. Governance fields are optional metadata. Memory is an external service. Lineage is not tracked.

This means two agents from different frameworks cannot interoperate structurally. No validator can determine whether an agent is well-formed. No governance system can verify compliance by inspecting the object itself. The agent is whatever the runtime says it is.

The cognition-compatible agent schema defines a canonical six-field object where every agent — full or partial — is structurally validated against typed fields for governance, memory, lineage, execution eligibility, identity, and semantic role. If the fields are absent or malformed, the object is not an agent. Governance is not applied to agents. Governance is what makes something an agent.

Interoperable, validatable, governable

A typed schema means agents can be validated at rest, not just at runtime. A governance system can inspect an agent object, verify its fields, check its lineage chain, and determine its execution eligibility without running it. Partial agents — objects that carry some but not all fields — can participate in the system with appropriately restricted roles.

This is the structural foundation that every other capability in the architecture depends on. Inference control gates semantic transitions on agent state. Confidence governance reads agent fields to determine execution readiness. Cryptographic governance signs and verifies agent policy fields. Without a canonical schema, none of these systems can operate on a common object. The schema is the interoperability layer.