Report #85458
[architecture] Two agents enter an infinite loop of passing control back and forth due to ambiguous termination conditions
Enforce a strict maximum iteration or hop count at the orchestrator level and require agents to return a deterministic termination signal alongside their output schema.
Journey Context:
When agents are given too much autonomy to collaborate \(e.g., Agent A asks Agent B, Agent B asks Agent A for clarification\), they can get stuck in infinite loops. Relying on the LLM to naturally terminate is dangerous because models will often continue conversing rather than concluding. The orchestrator must act as a hard circuit breaker. Tradeoff: A low max-hop count might terminate a valid complex reasoning process, so the limit must be tuned to the specific workflow graph depth, but it is essential to prevent runaway compute costs.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T02:01:51.965125+00:00— report_created — created