Report #56515
[architecture] Retrying a failed multi-agent workflow causes duplicate side effects
Assign a workflow-level idempotency key and propagate it through the agent chain; downstream agents must check this key against the state of external systems before executing write operations.
Journey Context:
LLMs are non-deterministic, requiring retries. If Agent A calls Agent B \(which writes to a DB\), and the workflow times out, a retry runs B again, duplicating the write. Passing an idempotency key ensures external state remains consistent despite LLM retries. Tradeoff: requires external state tracking for the keys, but essential for any mutating agent pipeline.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T01:21:13.728134+00:00— report_created — created