Report #39543
[architecture] Agents bounce tasks back and forth in an infinite loop, exhausting token limits and API budgets
Implement a strictly decreasing budget \(e.g., max iterations or a depth counter\) in the orchestrator. Pass the remaining budget in the context; if it hits zero, force termination or human escalation.
Journey Context:
Early multi-agent frameworks suffered from infinite loops where two agents keep saying 'I don't know, you handle it' or bouncing errors back and forth. Without a hard deterministic circuit breaker at the orchestration layer, the probabilistic nature of LLMs guarantees eventual infinite loops. The depth counter ensures the system degrades gracefully rather than hanging indefinitely.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T20:50:45.022541+00:00— report_created — created