LLM-Driven Collaborative Model for Untangling Commits via Explicit and Implicit Dependency Reasoning

Jul 16, 2026

Upload by Himbo

Atomic commits represent single development concerns, yet many real repositories contain tangled commits that mix unrelated changes, making code review and maintenance harder. Prior untangling methods rely on shallow signals and struggle to separate explicit dependencies (control/data flow) from implicit ones (semantic or conceptual links). ColaUntangle introduces a collaborative multi-agent framework using LLM-driven agents to model both dependency types, iterating through consultation and synthesizing perspectives. Evaluated on C# and Java datasets, it achieves 44% and 82% improvements over the best baseline.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages29

Reading Time1 h 13 min

Introduction

Atomic commits and their benefits

Tangled commits and their causes

Related work and limitations

ColaUntangle Approach

Multi-agent collaborative framework

Explicit vs implicit dependency modeling

Explicit and implicit contexts construction

Experimental Evaluation

Datasets

Results and comparison with baselines

Conclusion

arXiv :2507 . 16395v 3 [ cs .AI] 13 Jul 2026 LLM-Driven Collaborative Model for Untangling Commits via Explicit and Implicit Dependency Reasoning BO HOU, State Key Laboratory of Complex & Critical Software Environment (SKLCCSE), School of Computer Science and Engineering, Beihang University, China XIN TAN∗ , SKLCCSE, School of Computer Science and Engineering, Beihang University, China KAI ZHENG, SKLCCSE, School of Computer Science and Engineering, Beihang University, China FANG LIU, SKLCCSE, School of Computer Science and Engineering, Beihang University, China YINGHAO ZHU, School of Computing and Data Science, The University of Hong Kong, China LI ZHANG, SKLCCSE, School of Computer Science and Engineering, Beihang University, China Atomic commits, which address a single development concern, are a best practice in software development. In practice, however, developers often produce tangled commits that mix unrelated changes, complicating code review and maintenance. Prior untangling approaches—rule-based, feature-based, or graph-based—have made progress but typically rely on shallow signals and struggle to distinguish explicit dependencies (e.g., control/data flow) from implicit ones (e.g., semantic or conceptual relationships) . In this paper, we propose ColaUntangle, a new collaborative consultation framework for commit untangling that models both explicit and implicit dependencies among code changes. ColaUntangle integrates Large Language Model (LLM)-driven agents in a multi-agent architecture: one agent specializes in explicit dependencies, another in implicit ones, and a reviewer agent synthesizes their perspectives through iterative consultation. To capture structural and contextual information, we construct Explicit and Implicit Contexts, enabling agents to reason over code relationships with both symbolic and semantic depth. We evaluate ColaUntangle on two widely-used datasets (1,612 C\# and 14k Java tangled commits) . Experimental results show that ColaUntangle outperforms the best-performing baseline, achieving an improvement of 44% on the C\# dataset and 82% on the Java dataset. These findings highlight the potential of LLM-based collaborative frameworks for advancing automated commit untangling tasks. CCS Concepts: • Software and its engineering → Automatic programming; Software configuration management and version control systems. Additional Key Words and Phrases: Commit untangling, Explicit and implicit dependencies, LLMs, Multi-agent collaboration ACM Reference Format: Bo Hou, Xin Tan, Kai Zheng, Fang Liu, Yinghao Zhu, and Li Zhang. 2018. LLM-Driven Collaborative Model for Untangling Commits via Explicit and Implicit Dependency Reasoning. J. ACM 37, 4, Article 111 (August 2018), 29 pages. [https://doi.org/XXXXXXX.XXXXXXX](https://doi.org/XXXXXXX.XXXXXXX) ∗ Corresponding author. Authors’ Contact Information: Bo Hou, State Key Laboratory of Complex & Critical Software Environment (SKLCCSE), School of Computer Science and Engineering, Beihang University, Beijing, China, [houbo2024@buaa.edu.cn](houbo2024@buaa.edu.cn); Xin Tan, SKLCCSE, School of Computer Science and Engineering, Beihang University, Beijing, China, [xintan@buaa.edu.cn](xintan@buaa.edu.cn); Kai Zheng, SKLCCSE, School of Computer Science and Engineering, Beihang University, Beijing, China, [21373443@buaa.edu.cn](21373443@buaa.edu.cn); Fang Liu, SKLCCSE, School of Computer Science and Engineering, Beihang University, Beijing, China, [fangliu@buaa.edu.cn](fangliu@buaa.edu.cn); Yinghao Zhu, School of Computing and Data Science, The University of Hong Kong, Hong Kong SAR, China, [zhuyinghao@buaa.edu.cn](zhuyinghao@buaa.edu.cn); Li Zhang, SKLCCSE, School of Computer Science and Engineering, Beihang University, Beijing, China, [lily@buaa.edu.cn](lily@buaa.edu.cn). Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for

FAQs

What performance improvements does ColaUntangle achieve?

Across two widely-used datasets, ColaUntangle improves by 44% on the C# dataset and 82% on the Java dataset compared with the best-performing baseline.

Jul 16, 2026

Upload by Himbo