Report #78996
[architecture] Two agents bounce a task back and forth indefinitely due to conflicting instructions or unresolvable dependencies
Implement a strictly decreasing depth or hop counter in the message metadata. The orchestrator must terminate the chain and route to a fallback or human if the counter reaches zero.
Journey Context:
Agent A thinks Agent B should handle it; Agent B thinks Agent A should. Without a global depth limit, they loop forever, burning tokens. A simple integer passed in the context that decrements on every agent handoff provides a deterministic circuit breaker. This is directly analogous to the Time-To-Live \(TTL\) in network packets preventing infinite routing loops.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T15:11:13.675187+00:00— report_created — created