arXiv subjects

Lingxiao Jiang

Publications and source records attributed to Lingxiao Jiang.

At least 19 recordsLinked to original sources

arXiv2026

Better Understanding, Better Fixes? A Study of Hallucination in LLM-based Automated Program Repair

Large language models (LLMs) have significantly advanced automated program repair (APR), yet existing evaluations remain largely result-centric and provide limited insight into hallucination during repair. In APR, hallucination may arise not only in final patches but also in the intermediate artifacts that guide patch generation. To address this gap, we perform a multi-layered analysis of hallucination throughout the APR process. Specifically, we characterize hallucination as the production of patches or intermediate artifacts that are not faithfully grounded in the available repair evidence. We examine repair hallucination in final patches and understanding hallucination in intermediate artifacts through three tasks, namely triggering testcase identification, line coverage prediction, and additional testcase generation.We then evaluate three representative LLMs on 832 Defects4J bugs through automatic evaluation and manual analysis. Our results show that both repair and understanding hallucinations remain prevalent. Across models and settings, only 21.0%-55.9% of generated patches pass the developer-written test suite. Moreover, although more accurate intermediate artifacts are generally associated with successful repairs, this relationship does not always hold. Manual analysis of 812 sampled repairs identifies repair hallucinations in 72.7% of cases, including patches that pass all available tests; incorrect causal localization and incorrect repair strategies account for 45.9% and 18.5% of these hallucinations, respectively. Meanwhile, models frequently misidentify triggering testcases, mispredict line coverage involving branching control flow, and generate additional testcases with missing bug-triggering conditions or incorrect expected behavior.

cs.SE↗

arXiv2026

Improving Code Understanding in Large Language Models through Concept-Aware Consistency Learning

Large language models (LLMs) have recently shown impressive results on diverse code-related tasks, benefiting from large-scale training and instruction tuning. However, studies reveal that their grasp of fundamental programming concepts, such as data flow and control flow, remains shallow, leading to fragile performance when code requires deeper reasoning. This limitation restricts the practical adoption of LLMs in real-world software development. To address this issue, this work introduces a counterfactual code augmentation framework combined with concept-aware tuning, designed to guide LLMs toward stronger conceptual understanding. Comprehensive evaluation across multiple models and benchmarks demonstrates the effectiveness of the proposed approach.

cs.SE↗

arXiv2026

Are Performance-Optimization Benchmarks Reliably Measuring Coding Agents?

Repository-level performance-optimization benchmarks such as GSO, SWE-Perf and SWE-fficiency evaluate coding agents by applying patches to real repositories and comparing runtime against unoptimized baselines and official reference patches. Their leaderboard scores are increasingly used as evidence of coding-agent progress, but those scores can conflate runtime instability, benchmark-specific scoring rules, and how many tasks are already solved by at least one public submission. We audit these issues across the three benchmarks. First, we replay the official reference patches for 740 code optimization tasks across four common types of Google Cloud machines. Most benchmark tasks can be replayed, but their reference patches satisfy the original benchmark validity rules in every cross-machine replay for only 39/102 GSO tasks, 11/140 SWE-Perf tasks, and 411/498 SWE-fficiency tasks; SWE-Perf is especially fragile because many reference patches produce close-to-zero runtime changes. Second, we show that public submission rankings depend strongly on the benchmark scoring rule. Among eight public submissions shared by GSO and SWE-fficiency, the official rankings disagree on 9 of 28 pairwise submission comparisons, and SWE-fficiency's leaderboard scoring rule assigns the worst ten tasks overly high score weights of 58.5%-82.8%. Third, looking across 10 public submissions for each task, we find that at least one submission matches or beats the reference patch on 85.3% (384/450) of replay-valid GSO and SWE-fficiency tasks, and beats the unoptimized base code on 99.8% (449/450). Our study complements leaderboard scores by identifying tasks with more reliable performance signals, quantifying per-task score contributions, and exposing the remaining performance gaps that are hidden by aggregate rankings.

cs.SE↗

arXiv2026

Exploration Structure in LLM Agents for Multi-File Change Localization

Software engineering tools increasingly rely on LLM based agents to localize files to change to resolve a software issue. Most AI agents explore repositories linearly, that is, visiting one directory or file per step. We postulate that this is a structural mismatch for changes that span several subsystems. We compare linear sequential exploration against non-linear, domain-scoped parallel agentic exploration. Using SWE Bench Pro as initial benchmark, we focus on ansible as an exemplar. We construct an approach for persistent-session evaluation of GitHub issues anchored at a single base commit. We compare our non-linear domain-agent file traversal system against a base LLM without direct repository access, a single agent Recursive Language Model (RLM) baseline with a persistent Python REPL and an external CLI baseline using Codex 5.5 High. Domain scoped parallel agent spawning with a small Haiku-class model achieves the highest micro F1 among Haiku class models by a large margin. Domain-agents is the second highest behind only the much larger Codex 5.5 High on our own expanded benchmark including over more recent PRs from 2025 and 2026. On the original, curated, 2020 SWE-bench Pro benchmark, a larger Sonnet plain LLM baseline attains higher micro F1 by predicting few files, leading to higher precision, but at significantly lower all gold recall. We also present three additional findings. First, documentation evolution is a latent dependency unresolved by any approach. Second, naive file system access can degrade localization driven by test-file over prediction. Lastly, forced multi-agent consultation does not measurably help and raises token cost substantially.

cs.SE↗

arXiv2026

Exploring Code Analysis: Zero-Shot Insights on Syntax and Semantics with LLMs

Code analysis is fundamental in Software Engineering, supporting debugging, optimization, and security assessment. Human developers approach it through syntax parsing, static semantics inference, and dynamic reasoning. Traditional tools are effective but limited by language specificity and weak cross-language generalization. Large language models (LLMs) are promising for code tasks, yet their capabilities for fundamental code analysis remain underexplored. We structure our study around three aspects aligned with human practices: syntax parsing, static semantics inference, and dynamic reasoning. We evaluate 21 state-of-the-art LLMs across nine tasks in four languages (C, Java, Python, Solidity), including AST generation, CFG construction, data dependency, taint analysis, and flaky test reasoning. We apply a three-layer evaluation protocol (automated metrics, expert adjudication, consistency validation) to 3,124 code samples, achieving high inter-rater reliability (Cohen's kappa = 0.844-0.936) and strong human-machine agreement (Gwet's AC1 = 0.500-0.727, F1 = 0.791-0.882). While the best LLMs excel in syntax parsing (AST 90%+, expression matching 84-100%) and show promise in static analysis, their dynamic reasoning remains limited (<70%) with high data-shift sensitivity (per-project F1 varying 0-1.0). This hierarchy holds across model families and scales, suggesting fundamental rather than transient limitations. These findings show how LLMs complement traditional analyzers: they offer cross-language generalization but non-deterministic outputs needing validation, while traditional tools give deterministic guarantees but need language-specific configuration. We contribute a validated evaluation framework with comparison against traditional analyzers (Tree-sitter, Soot, Joern) and task-specific applicability tiers. Benchmark: https://github.com/mathieu0905/llm_code_analysis.git

cs.SE↗

arXiv2026

Same Signal, Different Semantics: A Cross-Framework Behavioral Analysis of Software Engineering Agents

Behavioral studies of LLM-based software engineering agents extract operational rules about which trajectory shapes correlate with higher resolution rates: that a test step follows a code modification, that error cascades are short, or that trajectories are compact. Each rule is typically derived from a single framework, and whether it transfers, in sign as well as magnitude, to structurally different agent designs has not been directly tested. We address this at ecosystem scale: 64,380 SWE-bench runs from 126 agent configurations spanning 43 frameworks, where each configuration pairs an LLM with a framework (e.g., SWE-Agent, OpenHands) that supplies its tools and workflow. We separate framework effects from LLM effects by holding each layer fixed in turn, then measure one behavior-outcome effect per configuration and examine how those effects agree or disagree. Swapping the framework while the LLM is held fixed produces large behavioral differences in every action feature. On most signals, configurations disagree not merely in magnitude but in direction. Error rate is the cleanest case: 47 configurations resolve more issues when their error rate is lower, while 48 resolve more when it is higher. Five other continuous features and three of seven binary patterns from prior SE literature show similar directional disagreement. Framework identity accounts for more of this variation than LLM family: for mean turns, framework explains 64% of the between-configuration variance against the LLM's 10%. The implication is that the same observable behavioral signal can carry opposite meaning for different agent configurations. Behavioral findings from any single framework therefore warrant cross-configuration validation before being claimed as general.

cs.SE↗

arXiv2026

Beyond Final Code: A Process-Oriented Error Analysis of Software Development Agents in Real-World GitHub Scenarios

AI-driven software development has rapidly advanced with the emergence of software development agents that leverage large language models (LLMs) to tackle complex, repository-level software engineering tasks. These agents go beyond just generation of final code; they engage in multi-step reasoning, utilize various tools for code modification and debugging, and interact with execution environments to diagnose and iteratively resolve issues. However, most existing evaluations focus primarily on static analyses of final code outputs, yielding limited insights into the agents' dynamic problem-solving processes. To fill this gap, we conduct an in-depth empirical study on 3,977 solving-phase trajectories and 3,931 testing-phase logs from 8 top-ranked agents evaluated on 500 GitHub issues in the SWE-Bench benchmark. Our exploratory analysis shows that Python execution errors during the issue resolution phase correlate with lower resolution rates and increased reasoning overheads. We have identified the most prevalent errors -- such as ModuleNotFoundError and TypeError -- and highlighted particularly challenging errors like OSError and database-related issues (e.g., IntegrityError) that demand significantly more debugging effort. Furthermore, we have discovered 3 bugs in the SWE-Bench platform that affect benchmark fairness and accuracy; these issues have been reported to and confirmed by the maintainers. To promote transparency and foster future research, we publicly share our datasets and analysis scripts.

cs.SE↗

arXiv2026

Rethinking the Value of Agent-Generated Tests for LLM-Based Software Engineering Agents

Large Language Model (LLM) code agents increasingly resolve repository-level issues by iteratively editing code, invoking tools, and validating candidate patches. In these workflows, agents often write tests on the fly, but the value of this behavior remains unclear. For example, GPT-5.2 writes almost no new tests yet achieves performance comparable to top-ranking agents.This raises a central question: do such tests meaningfully improve issue resolution, or do they mainly mimic a familiar software-development practice while consuming interaction budget? To better understand the role of agent-written tests, we analyze trajectories produced by six strong LLMs on SWE-bench Verified. Our results show that test writing is common, but resolved and unresolved tasks within the same model exhibit similar test-writing frequencies. When tests are written, they mainly serve as observational feedback channels, with value-revealing print statements appearing much more often than assertion-based checks. Based on these insights, we perform a prompt-intervention study by revising the prompts used with four models to either increase or reduce test writing. The results suggest that prompt-induced changes in the volume of agent-written tests do not significantly change final outcomes in this setting. Taken together, these results suggest that current agent-written testing practices reshape process and cost more than final task outcomes.

cs.SE↗

arXiv2026

AgentGuard: A Multi-Agent Framework for Robust Package Confusion Detection via Hybrid Search and Metadata-Content Fusion

The proliferation of open-source software (OSS) has made software supply chains prime targets for attacks like Package Confusion, where adversaries publish malicious packages with names deceptively similar to legitimate ones. To protect against such attacks and safeguard the use of OSS, multiple confusion detection methods have been proposed. However, existing methods are limited to single-signal retrieval strategies (relying solely on lexical or semantic metrics), struggle with high false positive rates (FPR), and are vulnerable to adversarial evasion. Critically, as content-agnostic approaches, they fundamentally fail to distinguish benign packages with high naming similarity from malicious, code-dissimilar impersonations, leading to persistent high FPR. To address these limitations, we introduce AgentGuard, a novel multi-agents based framework for package confusion detection. Specifically, it first discovers potential confusion targets using fine-tuned word embedding models with hybrid similarity search. After that, It subsequently evaluates risk via a fused machine learning model that uniquely combines: (1) a multi-dimensional metadata group and (2) a novel package content analysis group, to reduce the FPR and mitigate the impact of adversarial evasion. To assess the effectiveness of AgentGuard, we evaluate it on challenging ConfuDB and NeupaneDB datasets. Our results demonstrate that AgentGuard significantly outperforms state-of-the-art baselines, ConfuGuard and Typomind, improving precision by 12\%-49\% while simultaneously reducing the FPR by 11\%-35\%, and effectively discovers the confused package.

cs.SE↗

arXiv2025

Vital: Vulnerability-Oriented Symbolic Execution via Type-Unsafe Pointer-Guided Monte Carlo Tree Search

How to find memory safety bugs efficiently when navigating a symbolic execution tree that suffers from path explosion? Existing solutions either adopt path search heuristics to maximize coverage rate or chopped symbolic execution to skip uninteresting code (i.e., manually labeled as vulnerability-unrelated) during path exploration. However, most existing search heuristics are not vulnerability-oriented, and manual labeling of irrelevant code-to-be-skipped relies heavily on prior expert knowledge, making it hard to detect vulnerabilities effectively in practice. This paper proposes Vital, a new vulnerability-oriented path exploration for symbolic execution with two innovations. First, a new indicator (i.e., type-unsafe pointers) is suggested to approximate vulnerable paths. A pointer that is type-unsafe cannot be statically proven to be safely dereferenced without memory corruption. Our key hypothesis is that a path with more type-unsafe pointers is more likely to be vulnerable. Second, a new type-unsafe pointer-guided Monte Carlo Tree Search algorithm is implemented to guide the path exploration towards the areas that contain more unsafe pointers, aiming to increase the likelihood of detecting vulnerabilities. We built Vital on top of KLEE and compared it with existing path searching strategies and chopped symbolic execution. In the former, the results demonstrate that Vital could cover up to 90.03% more unsafe pointers and detect up to 57.14% more unique memory errors. In the latter, the results show that Vital could achieve a speedup of up to 30x execution time and a reduction of up to 20x memory consumption to detect known vulnerabilities without prior expert knowledge automatically. In practice, Vital also detected one previously unknown vulnerability (a new CVE ID is assigned), which has been fixed by developers.

cs.SE↗

arXiv2025

Cottontail: Large Language Model-Driven Concolic Execution for Highly Structured Test Input Generation

How can we perform concolic execution to generate highly structured test inputs for systematically testing parsing programs? Existing concolic execution engines are significantly restricted by (1) input structure-agnostic path constraint selection, leading to the waste of testing effort or missing coverage; (2) limited constraint-solving capability, yielding many syntactically invalid test inputs; (3) reliance on manual acquisition of highly structured seed inputs, resulting in non-continuous testing. This paper proposes Cottontail, a new Large Language Model (LLM)-driven concolic execution engine, to mitigate the above limitations. A more complete program path representation, named Expressive Structural Coverage Tree (ESCT), is first constructed to select structure-aware path constraints. Later, an LLM-driven constraint solver based on a Solve-Complete paradigm is designed to solve the path constraints smartly to get test inputs that are not only satisfiable to the constraints but also valid to the input syntax. Finally, a history-guided seed acquisition is employed to obtain new highly structured test inputs either before testing starts or after testing is saturated. We implemented Cottontail on top of SymCC and evaluated eight extensively tested open-source libraries across four different formats (XML, SQL, JavaScript, and JSON). Cottontail significantly outperforms baseline approaches by 30.73% and 41.32% on average in terms of line and branch coverage. Besides, Cottontail found six previously unknown vulnerabilities (six CVEs assigned). We have reported these issues to developers, and four out of them have been fixed so far.

cs.SE↗

arXiv2025

Prompt Stability in Code LLMs: Measuring Sensitivity across Emotion- and Personality-Driven Variations

Code generation models are widely used in software development, yet their sensitivity to prompt phrasing remains under-examined. Identical requirements expressed with different emotions or communication styles can yield divergent outputs, while most benchmarks emphasize only peak performance. We present PromptSE (Prompt Sensitivity Evaluation), a framework that creates semantically equivalent prompt variants with emotion and personality templates, and that evaluates stability using probability aware continuous scoring or using binary pass rates when logits are unavailable. The results are aggregated into a proposed area under curve metric (AUC-E) for cross model comparison. Across 14 models from three families (Llama, Qwen, and DeepSeek), our study shows that performance and stability behave as largely decoupled optimization objectives, and it reveals architectural and scale related patterns that challenge common assumptions about model robustness. The framework supports rapid screening for closed-source models as well as detailed stability analysis in research settings. PromptSE enables practitioners to quantify performance stability trade offs for deployment and model selection, positioning prompt stability as a complementary evaluation dimension alongside performance and fairness, and contributing to more trustworthy AI-assisted software development tools.

cs.SE↗

arXiv2025

VulnRepairEval: An Exploit-Based Evaluation Framework for Assessing Large Language Model Vulnerability Repair Capabilities

The adoption of Large Language Models (LLMs) for automated software vulnerability patching has shown promising outcomes on carefully curated evaluation sets. Nevertheless, existing datasets predominantly rely on superficial validation methods rather than exploit-based verification, leading to overestimated performance in security-sensitive applications. This paper introduces VulnRepairEval, an evaluation framework anchored in functional Proof-of-Concept (PoC) exploits. Our framework delivers a comprehensive, containerized evaluation pipeline that enables reproducible differential assessment, where repair success requires the original exploit to fail execution against the modified code. The benchmark construction involved extensive data curation: we processed over 400 CVEs and approximately 2,500 potential sources to extract a collection of authentic vulnerability instances (23 Python CVEs) amenable to automated testing with working PoCs. Through VulnRepairEval, we conduct a comprehensive evaluation of 12 popular LLMs and observe a significant performance deficit: even the top-performing model successfully addresses merely 5/23 instances (about 21.7%), exposing critical weaknesses in security-focused applications. Our failure analysis reveals that most unsuccessful attempts stem from imprecise vulnerability identification and patches containing syntactic or semantic errors. Enhanced prompting strategies and multi-agent approaches yield minimal improvements, with overall effectiveness remaining largely unaffected. This work contributes a stringent, practical evaluation framework for LLM-driven vulnerability remediation and underscores the necessity for assessment protocols that authentically reflect real-world exploitation scenarios.

cs.SE↗

arXiv2025

Importance Sampling is All You Need: Predict LLM's performance on new benchmark by reusing existing benchmark

With the rapid advancement of large language models , code generation has become a key benchmark for evaluating LLM capabilities. However, existing benchmarks face two major challenges: (1) the escalating cost of constructing high-quality test suites and reference solutions, and (2) the increasing risk of data contamination, which undermines the reliability of benchmark-based evaluations. In this paper, we propose BIS, a prompt-centric evaluation framework that enables ground-truth-free prediction of LLM performance on code generation tasks. Rather than executing generated code, BIS estimates performance metrics by analyzing the prompt distribution alone. Built on importance sampling theory and implemented using Importance Weighted Autoencoders, our method reweights samples from existing annotated benchmarks to estimate performance on new, unseen benchmarks. To stabilize the estimation, we introduce weight truncation strategies and compute marginal expectations across the fitted distributions. BIS serves as a complementary tool that supports benchmark development and validation under constrained resources, offering actionable and quick feedback for prompt selection and contamination assessment. We conduct extensive experiments involving 8,000 evaluation points across 4 CodeLlama models and 9 diverse benchmarks. Our framework achieves an average absolute prediction error of 1.1% for code correctness scores, with best- and worst-case errors of 0.3% and 1.9%, respectively. It also generalizes well to other metrics, attaining average absolute errors of 2.15% for pass@1. These results demonstrate the reliability and broad applicability of BIS, which can significantly reduce the cost and effort of benchmarking LLMs in code-related tasks.

cs.AI↗

arXiv2025

Runtime Anomaly Detection for Drones: An Integrated Rule-Mining and Unsupervised-Learning Approach

UAVs, commonly referred to as drones, have witnessed a remarkable surge in popularity due to their versatile applications. These cyber-physical systems depend on multiple sensor inputs, such as cameras, GPS receivers, accelerometers, and gyroscopes, with faults potentially leading to physical instability and serious safety concerns. To mitigate such risks, anomaly detection has emerged as a crucial safeguarding mechanism, capable of identifying the physical manifestations of emerging issues and allowing operators to take preemptive action at runtime. Recent anomaly detection methods based on LSTM neural networks have shown promising results, but three challenges persist: the need for models that can generalise across the diverse mission profiles of drones; the need for interpretability, enabling operators to understand the nature of detected problems; and the need for capturing domain knowledge that is difficult to infer solely from log data. Motivated by these challenges, this paper introduces RADD, an integrated approach to anomaly detection in drones that combines rule mining and unsupervised learning. In particular, we leverage rules (or invariants) to capture expected relationships between sensors and actuators during missions, and utilise unsupervised learning techniques to cover more subtle relationships that the rules may have missed. We implement this approach using the ArduPilot drone software in the Gazebo simulator, utilising 44 rules derived across the main phases of drone missions, in conjunction with an ensemble of five unsupervised learning models. We find that our integrated approach successfully detects 93.84% of anomalies over six types of faults with a low false positive rate (2.33%), and can be deployed effectively at runtime. Furthermore, RADD outperforms a state-of-the-art LSTM-based method in detecting the different types of faults evaluated in our study.

cs.SE↗

arXiv2025

Adapting Knowledge Prompt Tuning for Enhanced Automated Program Repair

Automated Program Repair (APR) aims to enhance software reliability by automatically generating bug-fixing patches. Recent work has improved the state-of-the-art of APR by fine-tuning pre-trained large language models (LLMs), such as CodeT5, for APR. However, the effectiveness of fine-tuning becomes weakened in data scarcity scenarios, and data scarcity can be a common issue in practice, limiting fine-tuning performance. To alleviate this limitation, this paper adapts prompt tuning for enhanced APR and conducts a comprehensive study to evaluate its effectiveness in data scarcity scenarios, using three LLMs of different sizes and six diverse datasets across four programming languages. Prompt tuning rewrites the input to a model by adding extra prompt tokens and tunes both the model and the prompts on a small dataset. These tokens provide task-specific knowledge that can improve the model for APR, which is especially critical in data scarcity scenarios. Moreover, domain knowledge has proven crucial in many code intelligence tasks, but existing studies fail to leverage domain knowledge during the prompt tuning for APR. To close this gap, we introduce knowledge prompt tuning, an approach that adapts prompt tuning with six distinct types of code- or bug-related domain knowledge for APR. Our work, to the best of our knowledge, is the first to adapt and evaluate prompt tuning and the effectiveness of code- or bug-related domain knowledge for APR, particularly under data scarcity settings. Our evaluation results demonstrate that prompt tuning with knowledge generally outperforms fine-tuning under various experimental settings, achieving an average improvement of 87.33% over fine-tuning in data scarcity scenarios.

cs.SE↗

arXiv2025

RustMap: Towards Project-Scale C-to-Rust Migration via Program Analysis and LLM

Migrating existing C programs into Rust is increasingly desired, as Rust offers superior memory safety while maintaining C's high performance. However, vastly different features between C and Rust--e.g., distinct definitions and usages of pointers and references--pose significant challenges beyond mere syntactic translation. Existing automated translation tools, such as C2Rust, may rely too much on syntactic, template-based translation and generate unsafe Rust code that is hard for human developers to read, maintain, or even compile. More semantic-aware translation that produces safer, idiomatic, and runnable Rust code is much needed. This paper introduces a novel dependency-guided and large language model (LLM)-based C-to-Rust translation approach, RustMap, based on three key ideas: (1) Utilize LLM capabilities to produce idiomatic Rust code from given small pieces of C code, (2) Mitigate LLM limitations in handling large codebases by breaking project-scale C programs into smaller units for translation according to their usage dependencies and composing them into a runnable Rust program, and (3) Enhance the correctness of the translated Rust program by using test cases to check input/output equivalence, isolate faulty code when execution states deviate, and iteratively refine the translation using feedback from compilation and test errors. We empirically evaluate RustMap on 126 real-world programs, including 125 from Rosetta Code and a 7000+ line bzip2 implementation using GPT-4o as the LLM. RustMap shows promising results, guiding GPT-4o to produce idiomatic, readable, and functional Rust code with significantly less unsafe code than other tools, and revealing non-trivial translation patterns reusable for future research.

cs.SE↗

arXiv2024

Evaluating Software Development Agents: Patch Patterns, Code Quality, and Issue Complexity in Real-World GitHub Scenarios

In recent years, AI-based software engineering has progressed from pre-trained models to advanced agentic workflows, with Software Development Agents representing the next major leap. These agents, capable of reasoning, planning, and interacting with external environments, offer promising solutions to complex software engineering tasks. However, while much research has evaluated code generated by large language models (LLMs), comprehensive studies on agent-generated patches, particularly in real-world settings, are lacking. This study addresses that gap by evaluating 4,892 patches from 10 top-ranked agents on 500 real-world GitHub issues from SWE-Bench Verified, focusing on their impact on code quality. Our analysis shows no single agent dominated, with 170 issues unresolved, indicating room for improvement. Even for patches that passed unit tests and resolved issues, agents made different file and function modifications compared to the gold patches from repository developers, revealing limitations in the benchmark's test case coverage. Most agents maintained code reliability and security, avoiding new bugs or vulnerabilities; while some agents increased code complexity, many reduced code duplication and minimized code smells. Finally, agents performed better on simpler codebases, suggesting that breaking complex tasks into smaller sub-tasks could improve effectiveness. This study provides the first comprehensive evaluation of agent-generated patches on real-world GitHub issues, offering insights to advance AI-driven software development.

cs.SE↗