Report #8981
[architecture] Two agents bouncing a task back and forth indefinitely without resolution
Implement a global max-handoff counter in the orchestrator. If the counter exceeds the threshold, break the loop by routing to a fallback generalist agent or terminating with an error.
Journey Context:
LLMs lack global iteration awareness. Agent A might determine Agent B should handle a task, but Agent B lacks the context or capability and routes back to A. Without a circuit breaker, this loop continues until token limits or timeouts exhaust resources. A simple max-handoff counter is a trivially implementable but essential guardrail against this common multi-agent failure mode.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T07:05:34.120728+00:00— report_created — created