Report #36899
[architecture] Agents bounce control back and forth in an infinite loop due to overlapping responsibilities
Enforce a maximum depth or hop count in the agent routing metadata and explicitly define termination conditions that prevent re-delegation to the calling agent.
Journey Context:
Without a hardcoded max\_depth, agents can get stuck in 'I don't know, you do it' loops, endlessly passing the token. This is especially common when agent capabilities overlap. The tradeoff is that a hard limit might prematurely cut off valid deep reasoning chains, but it is strictly necessary to cap compute costs and prevent system hangs.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T16:24:38.676664+00:00— report_created — created