Multi-Agent Deadlocks: Speaker Selection Logic
Preventing 'Ping-Pong' loops in multi-agent orchestrations.
Steps
- Implement a 'Speaker History' stack to track agent transitions.
- Detect 'Agent A -> Agent B -> Agent A' patterns without state changes.
- Inject a 'Manager Override' turn after 4 identical transitions.
- Set a 'Max_Round_Robin' turn limit for specific sub-tasks.
- Force a 'Consensus Pass' if no resolution is reached in 5 turns.