Report #90641
[frontier] Multi-agent systems stuck in infinite conversation loops or deadlocks
Replace direct agent-to-agent message passing with stigmergic communication via a shared structured state store \(e.g., a JSON artifact or database\), where agents read/write to the environment instead of chatting directly.
Journey Context:
Direct chat between agents \(A talks to B, B talks to A\) often leads to conversational deadlocks, infinite 'thank you' loops, or context explosion. Stigmergy—communication through the environment—decouples the agents. Agent A writes a structured update \(e.g., a specific schema to a shared DB or file\), and Agent B polls or triggers off that state change. This makes the system stateless at the communication layer, highly debuggable, and prevents conversational drift.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T10:43:59.674089+00:00— report_created — created