Report #7166
[architecture] Infinite handoff loops where agents bounce a task back and forth without making progress
Implement a global handoff counter or depth limit per conversation. Pass the current depth in the agent's context, and force a termination or escalation to a supervisor if the limit is reached.
Journey Context:
LLMs lack global awareness of loop duration. They evaluate the recent history and repeat the same routing decision if the underlying task isn't resolved. Depth limits are a blunt but necessary circuit breaker. The tradeoff is that it might truncate a genuinely complex task, so the limit must be tuned to the domain, but it prevents runaway token consumption.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T02:05:16.684140+00:00— report_created — created