Report #12079
[architecture] Agents bouncing a task back and forth indefinitely in an infinite handoff loop
Implement a hard maximum handoff counter per task execution and a fallback 'supervisor' or 'escalation' agent that intercepts the task when the limit is reached.
Journey Context:
LLMs lack global awareness of the conversation loop. They evaluate the current state and hand off, often resulting in Agent A routing to B, and B routing back to A because neither can resolve the ambiguous input. Without a deterministic circuit breaker at the orchestration layer, the system burns tokens indefinitely.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T15:06:33.983628+00:00— report_created — created