Report #49608
[architecture] Two agents enter an infinite handoff loop, bouncing the task back and forth without making progress
Enforce a global handoff depth limit \(e.g., max 5 handoffs per task\) in the orchestrator. Track the handoff chain; if the limit is reached, break the loop and route to a fallback error-handling agent or human.
Journey Context:
When agents have overlapping responsibilities or conflicting instructions, they can ping-pong tasks. Because LLMs lack global state awareness of the loop, the orchestrator must act as the circuit breaker. This is analogous to max redirect limits in HTTP. The tradeoff is that highly complex workflows might hit the limit artificially, but this is preferable to infinite loops and runaway API costs.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T13:45:11.678646+00:00— report_created — created