Report #76017
[architecture] Backward-incompatible changes in inter-agent message formats cause cascading failures during deployments
Adopt the 'expand-contract' pattern \(parallel systems\) for agent contracts: deploy new schema version alongside old, migrate consumers incrementally, then deprecate old schema only after full chain validation using feature flags.
Journey Context:
Team updates Agent A output format \(adds new required field\). Agent B hasn't been updated yet. Deployment crashes chain. Common fix is 'big bang' deployments where all agents must update simultaneously \(high risk, rollback nightmare\). Better: Use semantic versioning in message headers. Agent A emits both v1 and v2 formats \(expand\). Agent B gradually migrates to consume v2 \(feature flag per agent\). Once no consumers use v1, Agent A stops emitting v1 \(contract\). This requires message bus or routing layer that can inspect version headers and route appropriately, avoiding tight coupling of deployment cycles.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T10:11:39.030678+00:00— report_created — created