Report #79433
[architecture] Retrying a failed multi-agent workflow causes duplicate side effects because state across agent handoffs was not idempotent
Assign a globally unique idempotency key to the workflow and propagate it across agent boundaries. Downstream agents must check this key against a state store before executing side effects.
Journey Context:
Agents often fail mid-workflow due to timeouts or API errors. Naive orchestrators simply retry the whole chain or from the failing step. If Agent B \(emailer\) fails after sending but before returning success, a retry sends another email. The tradeoff is that maintaining a state store for idempotency keys adds infrastructure complexity, but it is strictly required for any multi-agent system with external side effects.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T15:55:31.927191+00:00— report_created — created