Agent Declaration Layer · Control Surface

Authority before action.

Agent Authority is the control layer for agent-readable systems. It declares who may act, under whose delegation, with what scope, under what policy, in which jurisdiction, and subject to what revocation conditions.

An agent may be discoverable, identifiable, and technically capable while still lacking a legitimate right to act. The authority layer exists so a machine can evaluate that right before an irreversible action happens.

Identity answers who an agent is. Discovery answers where it can be reached. Capability answers what it can do. None of those answer whether it is authorized to do so.

The declaration layer note at agentjson.org and manifest-yaml.com already names that gap explicitly. This domain narrows the problem to control: delegation, policy, jurisdiction, and revocation.

The minimal example set already defines the control-side file family. On the authority side, the declared objects are:

/.well-known/agent.json The agent surface that points to authority, policy, jurisdiction, provenance, audit, and revocation objects.
/.well-known/policy.json The constraint layer: allowed actions, prohibited actions, and human-review requirements.
/.well-known/authority.json The delegation claim: who delegated, to whom, for what scope, under which limits.
/.well-known/jurisdiction.json The legal and geographic scope under which authority remains valid.
/.well-known/revocation.json The termination path: expiry, override, rescission, and policy supersession conditions.

These objects are not theory. They are already sketched concretely in the example set that accompanies AGENT_DECLARATION_LAYER_v0.1.

agent-authority.org is the control layer.

agent-provenance.org is the proof layer.

Authority answers whether an agent may act. Provenance answers how that action can later be defended, verified, and traced. One governs the permission boundary. The other governs the evidence boundary.