Request-Response Protocols

Request-response is the workhorse protocol — one agent asks, one agent answers. 73.41% of all inter-agent messages in our system follow this pattern. The protocol requires strict pairing: every request must receive a response within the TTL window, and every response must reference the original correlationId.

The failure modes are well-defined. Timeout: no response within TTL — the orchestrator retries once, then escalates. Rejection: the recipient cannot fulfill the request — the response includes a reason code and suggested alternative. Partial: the recipient can partially fulfill — the response includes what was completed and what remains. Each failure mode has a handler. Unhandled failures are the only unacceptable outcome. A message that disappears into the void with no response and no error is a system bug, not an edge case.