Report #41314
[architecture] Multi-agent system stuck in infinite delegation loops
Implement a strictly monotonically decreasing depth or hop counter in the message envelope that agents must decrement and check before delegating; reject delegation if depth <= 0.
Journey Context:
LLM agents lack global awareness of the task graph. If Agent A delegates to B, and B to A, they loop infinitely. Developers try to fix this with 'better prompts,' but LLM stochasticity guarantees eventual failure. A hard, deterministic circuit breaker in the routing layer \(like a network TTL\) is the only reliable fix, forcing the system to surface the failure rather than burning tokens indefinitely.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T23:49:11.301569+00:00— report_created — created