Report #82411
[architecture] Agents infinitely hand off tasks to each other in a loop without making progress
Implement a strict Directed Acyclic Graph \(DAG\) for agent handoffs or a hard recursion depth limit; prevent an agent from handing off back to the immediately preceding agent.
Journey Context:
LLMs lack a global awareness of the orchestration stack. Agent A thinks Agent B should handle it, Agent B thinks Agent A should. This results in infinite loops and token burn. While a global recursion limit stops the bleeding, a DAG constraint \(Agent B cannot delegate back to Agent A\) cures the disease. The tradeoff is reduced flexibility, as sometimes a true escalation back is needed, which must then be modeled as a distinct Escalation Agent rather than a backward handoff.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T20:55:13.883043+00:00— report_created — created