QueenBee Planner: Skill-Evolving Communication Topologies for Token-Efficient LLM Multi-Agent Systems

2026-06-25

AI Summary

This work addresses the limitations of fixed communication topologies or task-specific memory in multi-agent systems by modeling communication structures as retrievable, self-evolving skills. The proposed method employs an external LLM-based planner to generate temporal communication directed acyclic graphs (DAGs) that orchestrate information passing and fusion, while freezing worker agents and scoring functions to prevent interference. Innovatively treating communication topology as a learnable design skill, the approach integrates execution trajectory distillation, design rule extraction (Preserve/Modify/Avoid), held-out validation gating, variance-aware credit assignment, and motif-level attribution to mitigate misattribution and overfitting, thereby enabling transferable architectural knowledge accumulation. Experiments demonstrate significant improvements: on the Count-Frequency task, RMSE drops from 12.53 to 7.87, with marked reductions in message count, model invocations, and token consumption; superior performance is also achieved on Silo-Bench tasks compared to both fixed and cold-start topologies.

Abstract

Large language model (LLM) multi-agent systems increasingly depend not only on how individual agents reason, but also on how agents are connected. This paper introduces QueenBee Planner, a framework that treats inter-agent communication topology as a retrievable and self-improving design skill. A pool of worker agents, the task adapter, and the scoring function are frozen; only an outer LLM planner learns to generate temporal communication DAGs specifying who sends information to whom, in which round, who merges messages, and who emits the final answer. Execution traces are distilled into evidence-backed design rules with three actions: \emph{Preserve}, \emph{Modify}, and \emph{Avoid}. To prevent self-evolution from turning lucky runs or plausible but false explanations into policy, QueenBee uses held-out acceptance gates, variance-aware credit, motif-level attribution, transfer trust, insight falsification, and structural deduplication. We evaluate the method on Count-Frequency aggregation and Silo-Bench-style distributed coordination tasks. With fixed workers, self-evolved graph generation produces communication structures that improve over fixed topologies and cold generation. In the CF fulltest setting, the best generated graph reduces RMSE from 12.53 for the strongest fixed topology to 7.87 while also reducing messages, model calls, and token cost; Silo-style results show the same direction of improvement over cold and fixed-topology baselines. These results suggest that multi-agent systems can learn reusable architectural design knowledge rather than merely memorizing task answers.

Problem

• communication topology
• multi-agent systems
• token efficiency
• LLM coordination
• graph generation

Innovation

• communication topology
• skill-evolving
• token-efficient
• multi-agent LLM
• self-improving design