StepFinder: A Temporal Semantic Framework for Failure Attribution in Multi-Agent Systems
StepFinder is a new framework designed for failure attribution in multi-agent systems. It aims to improve the identification of root causes for failures in these systems, which are often affected by execution errors. The framework demonstrates significant efficiency improvements over existing methods, reducing inference time by 79%.
- ▪StepFinder uses LLMs only during the feature construction phase to encode execution logs into temporal semantic sequences.
- ▪The framework combines temporal modeling and attention modules to capture dependencies in execution trajectories.
- ▪Experimental results show that StepFinder outperforms LLM-based methods in step-level failure attribution.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2606.03467 (cs) [Submitted on 2 Jun 2026] Title:StepFinder: A Temporal Semantic Framework for Failure Attribution in Multi-Agent Systems Authors:Taiyu Zhu, Yifan Wu, Weilin Jin, Ying Li, Gang Huang View a PDF of the paper titled StepFinder: A Temporal Semantic Framework for Failure Attribution in Multi-Agent Systems, by Taiyu Zhu and 3 other authors View PDF HTML (experimental) Abstract:LLM-based multi-agent systems exhibit remarkable collaborative capabilities in complex multi-step tasks. However, these systems are highly sensitive to single-step execution errors that can propagate through agent interactions and lead to cascading failures.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
