Report #21200
[architecture] Agent A delegates to Agent B, Agent B delegates back to Agent A, creating an infinite loop of handoffs
Enforce a maximum handoff depth or a strictly acyclic handoff graph. Track the handoff chain in the state and abort if a cycle is detected or depth limit is exceeded.
Journey Context:
LLMs do not naturally realize they are looping; they just see the latest prompt and think 'the other agent should handle this.' Without cycle detection, a multi-agent system will happily burn through tokens forever. The tradeoff of a strict depth limit is that deeply nested legitimate workflows might get cut off, so tracking the exact path and detecting cycles is strictly better than a flat depth limit.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T13:59:42.549171+00:00— report_created — created