Report #83211
[architecture] Retrying failed multi-agent workflows causes duplicate side effects \(emails, DB writes\)
Assign a unique, deterministic idempotency key \(e.g., workflow\_id \+ step\_id\) to every agent step that executes a state-mutating action, and check this key against a persistent ledger before executing the action.
Journey Context:
When an orchestrator retries a failed step, it often re-invokes the agent's tool calls. If the agent wrote to a database on the previous attempt but failed to return the result, the retry duplicates the action. Idempotency keys ensure exactly-once execution semantics, which is crucial for maintaining system consistency when agent steps are non-deterministic and prone to intermittent infrastructure failures.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T22:15:27.493476+00:00— report_created — created