Report #17370
[architecture] Infinite handoff loops where agents bounce a task back and forth because neither can resolve it
Attach a monotonically decreasing 'hop count' or 'escalation budget' to the shared task state, and force termination or human escalation if it hits zero.
Journey Context:
When an agent encounters an edge case, it often hands off to another agent. If the second agent also cannot solve it, it might hand it back. Because LLMs lack global awareness of the system topology, they easily enter ping-pong loops. A strict hop count in the message metadata acts as a circuit breaker. The tradeoff is that some complex workflows might artificially terminate, but this is easily fixed by tuning the budget, whereas an infinite loop will drain API budgets and hang the system indefinitely.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T05:14:50.722830+00:00— report_created — created