Agent-to-Relay protocol
Action Cards, Decision Receipts, decision factors, and trace in one message standard
Agent-to-Relay exchange
A proposed action becomes a clear Relay decision
Agent
Agent identity
Identity and capability refs attach to the proposed action
Source
Protocol
Action Card
Action, target, authority, evidence, policy, and risk use one shape
Request
Relay
Decision Receipt
Relay returns the decision, reason, next step, and trace
Response
Trace
Review reference
Request and decision stay reviewable without private payloads
Proof
What Protocol standardizes
The shared message objects every system can read
Action Card
Proposed action before execution
Agent to Relay
Decision Receipt
Decision, reason, next step, factors, and trace
Relay to Agent
Decision factors
Identity, authority, evidence, policy, and risk checks
Review basis
Review challenge
Missing condition, objection, or required restraint
Review input
Trace reference
Record for request, decision, and return
Review record
System boundaries
Separate Relay, Registry, and runtime roles
Registry supplies identity
Agent Passport refs give Relay identity and standing context
Identity source
Relay returns decisions
Relay reviews the Action Card and returns a Decision Receipt
Decision layer
Runtime keeps execution
Execution stays with the agent runtime
Execution owner
MCP stays optional
MCP can use the same Action Card and Decision Receipt language
Adapter path
Current public posture
Draft v0.1 is the current public reference
Status
Draft v0.1 communication protocol
Current draft
Canonical path
Reference defines objects, validation, and versioning
Cite /docs