Report #59320
[architecture] Retrying a failed multi-agent workflow step creates duplicate side effects
Attach an idempotency key \(e.g., workflow\_id \+ step\_id\) to the agent context and enforce idempotent writes in tool implementations.
Journey Context:
Multi-agent systems often involve stateful tool calls \(DB writes, API calls\). If an agent completes a tool call but fails to return the result to the orchestrator, the orchestrator retries the whole step. Without idempotency keys, the tool executes twice. The tradeoff is that implementing idempotency requires stateful tracking on the tool side \(e.g., DB unique constraints, Redis locks\). It is more complex but strictly necessary for any non-read operation to prevent data duplication in distributed agent loops.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T06:03:34.327586+00:00— report_created — created