SemAgent: A Semantics Aware Program Repair Agent

2025-06-19

AI Summary

Existing program repair agents suffer from insufficient semantic understanding, leading to overfitting and poor generalization due to excessive focus on suspicious code lines. This work proposes a semantics-driven, two-stage intelligent repair agent—the first to jointly integrate problem semantics (generalized abstract modeling), code semantics (context-sensitive understanding with multi-line collaborative isolation), and execution semantics (trace analysis and context-aware retrieval)—to establish a “repair–review” dual-phase decision workflow. The method significantly enhances robustness against multi-line bugs and edge-case failures. Evaluated on SWEBench-Lite, it achieves a 44.66% task resolution rate, outperforming the strongest baseline by 7.66 percentage points and surpassing all prior workflow-based approaches.

Technology Category

• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Large Language Models (LLMs) have shown impressive capabilities in downstream software engineering tasks such as Automated Program Repair (APR). In particular, there has been a lot of research on repository-level issue-resolution benchmarks such as SWE-Bench. Although there has been significant progress on this topic, we notice that in the process of solving such issues, existing agentic systems tend to hyper-localize on immediately suspicious lines of code and fix them in isolation, without a deeper understanding of the issue semantics, code semantics, or execution semantics. Consequently, many existing systems generate patches that overfit to the user issue, even when a more general fix is preferable. To address this limitation, we introduce SemAgent, a novel workflow-based procedure that leverages issue, code, and execution semantics to generate patches that are complete - identifying and fixing all lines relevant to the issue. We achieve this through a novel pipeline that (a) leverages execution semantics to retrieve relevant context, (b) comprehends issue-semantics via generalized abstraction, (c) isolates code-semantics within the context of this abstraction, and (d) leverages this understanding in a two-stage architecture: a repair stage that proposes fine-grained fixes, followed by a reviewer stage that filters relevant fixes based on the inferred issue-semantics. Our evaluations show that our methodology achieves a solve rate of 44.66% on the SWEBench-Lite benchmark beating all other workflow-based approaches, and an absolute improvement of 7.66% compared to our baseline, which lacks such deep semantic understanding. We note that our approach performs particularly well on issues requiring multi-line reasoning (and editing) and edge-case handling, suggesting that incorporating issue and code semantics into APR pipelines can lead to robust and semantically consistent repairs.

Problem

• Addresses hyper-localized fixes without deeper semantic understanding
• Improves patch generation to avoid overfitting to specific issues
• Enhances multi-line reasoning and edge-case handling in repairs

Innovation

• Leverages execution semantics for context retrieval
• Comprehends issue-semantics via generalized abstraction
• Two-stage architecture: repair and reviewer stages