ARISE: Graph-Based Tool for AI-Driven Bug Localization and Repair
The new system ARISE (Agentic Repository-level Issue Solving Engine) improves LLM-based agents for automating fault localization and program repair within software repositories. It features a multi-granularity program graph that enhances structural relationships to the statement level, linking nodes via intra-procedural definition-use edges for accurate function and line-level localization. This graph is made accessible through a three-tier tool API, enabling agents to track data-flow dependencies with a single call. ARISE overcomes the shortcomings of current graph-based systems, which typically only represent relationships at the file, class, and function levels without accounting for variable value flows within procedures. Details about ARISE can be found in a paper on arXiv (2605.03117v1).
Key facts
- ARISE stands for Agentic Repository-level Issue Solving Engine.
- It augments LLM-based agents with a multi-granularity program graph.
- The graph includes statement-level nodes and intra-procedural definition-use edges.
- ARISE provides a three-tier tool API for data-flow slicing.
- It targets repository-level fault localization and automated program repair.
- Existing graph-based systems lack semantic precision for function- and line-level localization.
- The paper is available on arXiv with ID 2605.03117v1.
Entities
Institutions
- arXiv