Report #69758
[architecture] Two agents get stuck in an infinite clarification loop burning tokens indefinitely
Embed a strictly decrementing TTL \(Time-To-Live\) or max-turn counter in the inter-agent message metadata. If the counter reaches zero, force an escalation or termination instead of allowing another agent handoff.
Journey Context:
LLMs can get stuck in polite but unproductive loops \(e.g., Agent A: 'I need more info', Agent B: 'Here is info', Agent A: 'Still not enough'\). Because LLMs lack global state awareness of how long they've been running, they cannot self-correct these loops. A deterministic TTL in the message header acts as a necessary circuit breaker.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T23:34:39.145199+00:00— report_created — created