Autonomous Repair for Multi-Agent Systems via Monte-Carlo Tree Search

2026-07-31

AI Summary

This work addresses the lack of automated repair mechanisms in multi-agent systems, which currently rely heavily on manual intervention. To bridge this gap, the study introduces Monte Carlo Tree Search (MCTS) into the domain of multi-agent system repair for the first time, proposing a diagnosis-guided node expansion strategy and a classification-augmented evaluation mechanism. Furthermore, it incorporates partial trajectory rollback to reduce computational overhead. The authors also construct StateMAS, the first large-scale, replayable benchmark for multi-agent system repair. Experimental results on StateMAS demonstrate that the proposed method improves repair success rates by 3.0%–12.1% over existing approaches while maintaining token consumption comparable to baseline methods.

Abstract

Multi-agent systems (MAS) are increasingly deployed to solve complex tasks. In case of incorrect or unsatisfactory outputs, users have to manually locate agent mistakes by inspecting agent trajectories (i.e., {\em failure attribution}) and provide feedback to refine the outputs (i.e., {\em repair}). Despite some recent work in MAS failure attribution, automated mechanisms to recover from such mistakes remain largely unexplored. To bridge this gap, we propose MARS, a search-based framework that formulates MAS repair as a Monte Carlo Tree Search (MCTS) process and navigates the vast space of potential repairs via diagnosis-guided expansion with taxonomy-augmented evaluation. Unlike standard MCTS, which evaluates a complete simulation via full rollout, MARS evaluates the agent trajectory using partial rollout to reduce token consumption. Furthermore, we introduce StateMAS, a large-scale MAS repair benchmark with 1,310 replayable multi-agent failure trajectories spanning four types of agent architectures and four LLM backbones. Experiments on StateMAS demonstrate that MARS consistently outperforms state-of-the-art methods, achieving an absolute improvement from 3.0\% to 12.1\% across all settings, while maintaining a comparable token consumption cost. The ablation study further confirms that taxonomy-augmented evaluation and diagnosis-guided expansion are critical to achieving these performance gains.

Problem

• multi-agent systems
• autonomous repair
• failure attribution
• Monte Carlo Tree Search
• agent trajectory

Innovation

• Monte Carlo Tree Search
• multi-agent repair
• diagnosis-guided expansion
• taxonomy-augmented evaluation
• partial rollout