Protocol for Governed Agent Action
Message structure for Action Cards, decision factors, receipts, and trace
Agent to Relay exchange
Protocol carries Action Cards into Relay receipts
Agent
Identity reference
Identity and capability refs travel with the Action Card
Attribution
Protocol
Action Card
Action, object, authority, evidence, policy, and risk become readable
Request
Relay
Decision Receipt
Relay returns decision factors, reason, next step, and trace
Response
Trace
Replayable reference
The exchange stays citeable without exposing payloads
Proof
What Protocol standardizes
Message objects and checks for the Action Card to Receipt path
Action Card
Agent request format before execution
Agent -> Relay
Decision Receipt
Relay response with decision, factors, reason, and trace
Relay -> Agent
Decision factors
Identity, authority, evidence, policy, and risk checks
Review basis
Challenge
Objection, missing condition, or restraint
Review input
Trace reference
Proof handle for request, decision, and return
Audit path
System boundaries
Protocol keeps each Neura role separate
Registry supports identity
Production Action Cards carry Agent Passport refs Relay can read
Lookup source
Relay owns decisions
Relay evaluates the Action Card and returns receipt, Registry status, and trace replay
Workspace proof
Agents own action intent
Agent runtime remains with the product
Participant layer
MCP stays optional
Sandbox and production MCP use the same Action Card and Decision Receipt language
Compatibility
Current public posture
Draft v0.1 is the launch Agent Relay reference
Status
Draft v0.1 communication protocol
Not stable release
Canonical path
Reference defines objects, validation, and versioning
Cite /docs