Report #53265
[architecture] Infinite loop between two agents delegating back to each other
Enforce a strictly acyclic delegation graph and a hard global recursion depth limit \(e.g., max 5 handoffs\) at the orchestrator level.
Journey Context:
In flexible multi-agent frameworks, agents can transfer conversations to each other. If Agent A doesn't know how to do X, it transfers to B. If B also fails, it might transfer back to A. Without cycle detection, this loops forever, burning tokens and hanging the system. The orchestrator must maintain a call stack and abort if a cycle is detected or max depth is reached. Tradeoff: Limits the problem-solving depth of the swarm, but prevents unbounded cost overruns and system hangs.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T19:54:16.697340+00:00— report_created — created