NeuraProtocol

Relay review message schema

Fields Relay needs to review a proposed action and return a traceable decision

ReferenceFlowReleases

Reference model

Draft v0.1 is explicit and boundary-aware

Agent identity

Who is acting

Agent and capability refs identify who is acting

Attribution

Proposed action

What is proposed

Type, target, authority, evidence, policy, risk, and outcome

Review input

Review factors

What Relay checks

Identity, authority, evidence, policy, and risk are checked

Review basis

Decision Receipt

What Relay returned

Decision, reason, confidence, next step, and trace come back

Review output

Protocol format

Concrete JSON for the active Draft v0.1 surface

Protocol format

Action Card and Decision Receipt JSON

Compact examples for request, response, and validation

Action Card
{
  "version": "0.1",
  "agent": {
    "id": "agent_ref:registry:customer_update_agent",
    "owner": "owner_ref:builder_workspace_01",
    "capability": "customer_update",
    "capabilityVersion": "capability_version_ref:v1"
  },
  "proposedAction": {
    "type": "email.send",
    "summary": "Send approved customer update email",
    "target": "customer_thread_ref:4821"
  },
  "affectedObject": "customer_thread_ref:4821",
  "context": {
    "evidenceRefs": [
      "approval_ref:manager_4821",
      "template_ref:customer_update_v3"
    ],
    "ruleRefs": [
      "policy_ref:customer_email_scope"
    ],
    "riskCategory": "customer_communication"
  },
  "requestedOutcome": "decision_receipt"
}

Operating rules

Protocol stays separate from Relay and execution

Readable by both sides

Agent and Relay read the same proposed action

Reviewable before execution

Warnings and blockers stay visible before anything changes