Gail Kaiser

Automated Program Repair (APR) agents leverage large language models (LLMs) to autonomously diagnose and patch software bugs using planning, reasoning, and tools. Although these agents show strong performance on leaderboards such as SWE-bench, little is understood about how they take actions, where they fail, and how their behavior compares to human developers. In this paper, we present the first systematic analysis of these limitations using 5 state-of-the-art APR agents. We trace the full decision-making pipelines of the 5 APR agents across 500 real-world repair tasks, from issue description to patch validation. Our study reveals that, while agents excel at simple fixes, they struggle with logic-intensive bugs, often generating verbose, overfitted patches that pass existing test suites without solving the root cause. Test generation and regression test selection remain major bottlenecks, as agents fail to reproduce issues or run relevant regression tests. Moreover, many agents operate with primitive tooling (e.g. bash scripts) and do not have access to debuggers or program analysis tools. These findings highlight key limitations of current APR systems and motivate several directions for next-generation APR design, including but not limited to: (1) a shift-left approach emphasizing early, high-quality test generation and validation to reduce spurious fixes and improve semantic correctness; (2) richer, more integrated tool ecosystems; (3) diversified agent architectures that combine complementary strengths; and (4) benchmarks that prioritize semantic repair quality and test-generation fidelity over surface-level success metrics.

With the advent of large language models (LLMs), software engineering agents (SWE agents) have emerged as a powerful paradigm for automating a range of software tasks -- from code generation and repair to test case synthesis. These agents operate autonomously by interpreting user input and responding to environmental feedback. While various agent architectures have demonstrated strong empirical performance, the internal decision-making worfklows that drive their behavior remain poorly understood. Deeper insight into these workflows hold promise for improving both agent reliability and efficiency. In this work, we present the first systematic study of SWE agent behavior through the lens of execution traces. Our contributions are as follows: (1) we propose the first taxonomy of decision-making pathways across five representative agents; (2) using this taxonomy, we identify three core components essential to agent success -- bug localization, patch generation, and reproduction test generation -- and study each in depth; (3) we study the impact of test generation on successful patch production; and analyze strategies that can lead to successful test generation; (4) we further conduct the first large-scale code clone analysis comparing agent-generated and developer-written patches and provide a qualitative study revealing structural and stylistic differences in patch content. Together, these findings offer novel insights into agent design and open avenues for building agents that are both more effective and more aligned with human development practices.

Pre-trained code language models have achieved promising performance in code generation and improved the programming efficiency of human developers. However, their self-refinement capability is typically overlooked by the existing evaluations of code LMs, which focus only on the accuracy of the one-time prediction. For the cases when code LMs fail to implement the correct program, developers actually find it hard to debug and fix the faulty prediction since it is not written by the developers themselves. Unfortunately, our study reveals that code LMs cannot efficiently self-refine their faulty generations as well. In this paper, we propose CYCLE framework, learning to self-refine the faulty generation according to the available feedback, such as the execution results reported by the test suites. We evaluate CYCLE on three popular code generation benchmarks, HumanEval, MBPP, and APPS. The results reveal that CYCLE successfully maintains, sometimes improves, the quality of one-time code generation, while significantly improving the self-refinement capability of code LMs. We implement four variants of CYCLE with varied numbers of parameters across 350M, 1B, 2B, and 3B, and the experiments show that CYCLE consistently boosts the code generation performance, by up to 63.5

This paper introduces a novel code-to-code search technique that enhances the performance of Large Language Models (LLMs) by including both static and dynamic features as well as utilizing both similar and dissimilar examples during training. We present the first-ever code search method that encodes dynamic runtime information during training without the need to execute either the corpus under search or the search query at inference time and the first code search technique that trains on both positive and negative reference samples. To validate the efficacy of our approach, we perform a set of studies demonstrating the capability of enhanced LLMs to perform cross-language code-to-code search. Our evaluation demonstrates that the effectiveness of our approach is consistent across various model architectures and programming languages. We outperform the state-of-the-art cross-language search tool by up to 44.7\%. Moreover, our ablation studies reveal that even a single positive and negative reference sample in the training process results in substantial performance improvements demonstrating both similar and dissimilar references are important parts of code search. Importantly, we show that enhanced well-crafted, fine-tuned models consistently outperform enhanced larger modern LLMs without fine tuning, even when enhancing the largest available LLMs highlighting the importance for open-sourced models. To ensure the reproducibility and extensibility of our research, we present an open-sourced implementation of our tool and training procedures called REINFOREST.

Code Large Language Models (Code LLMs) are being increasingly employed in real-life applications, so evaluating them is critical. While the conventional accuracy evaluates the performance of Code LLMs on a set of individual tasks, their self-consistency across different tasks is overlooked. Intuitively, a trustworthy model should be self-consistent when generating natural language specifications for its own code and generating code for its own specifications. Failure to preserve self-consistency reveals a lack of understanding of the shared semantics underlying natural language and programming language, and therefore undermines the trustworthiness of a model. In this paper, we first formally define the self-consistency of Code LLMs and then design a framework, IdentityChain, which effectively and efficiently evaluates the self-consistency and conventional accuracy of a model at the same time. We study eleven Code LLMs and show that they fail to preserve self-consistency, which is indeed a distinct aspect from conventional accuracy. Furthermore, we show that IdentityChain can be used as a model debugging tool to expose weaknesses of Code LLMs by demonstrating three major weaknesses that we identify in current models using IdentityChain. Our code is available at https://github.com/marcusm117/IdentityChain.

Most existing pre-trained language models for source code focus on learning the static code text, typically augmented with static code structures (abstract syntax tree, dependency graphs, etc.). However, program semantics will not be fully exposed before the real execution. Without an understanding of the program execution, statically pre-trained models fail to comprehensively capture the dynamic code properties, such as the branch coverage and the runtime variable values, and they are consequently less effective at code understanding tasks, such as retrieving semantic clones and detecting software vulnerabilities. To close the gap between the static nature of language models and the dynamic characteristics of programs, we introduce TRACED, an execution-aware pre-training strategy for source code. Specifically, we pre-train code language models with a combination of source code, executable inputs, and corresponding execution traces. Our goal is to teach code models the complicated execution logic during the pre-training, enabling the model to statically estimate the dynamic code properties without repeatedly executing code during task-specific fine-tuning. To illustrate the effectiveness of our proposed approach, we fine-tune and evaluate TRACED on three downstream tasks: static execution estimation, clone retrieval, and vulnerability detection. The empirical results show that TRACED relatively improves the statically pre-trained code models by 12.4% for complete execution path prediction and by 25.2% for runtime variable value predictions. TRACED also significantly outperforms statically pre-trained models in clone retrieval and vulnerability detection across four public benchmarks.

Self-driving cars and trucks, autonomous vehicles (AVs), should not be accepted by regulatory bodies and the public until they have much higher confidence in their safety and reliability -- which can most practically and convincingly be achieved by testing. But existing testing methods are inadequate for checking the end-to-end behaviors of AV controllers against complex, real-world corner cases involving interactions with multiple independent agents such as pedestrians and human-driven vehicles. While test-driving AVs on streets and highways fails to capture many rare events, existing simulation-based testing methods mainly focus on simple scenarios and do not scale well for complex driving situations that require sophisticated awareness of the surroundings. To address these limitations, we propose a new fuzz testing technique, called AutoFuzz, which can leverage widely-used AV simulators' API grammars to generate semantically and temporally valid complex driving scenarios (sequences of scenes). To efficiently search for traffic violations-inducing scenarios in a large search space, we propose a constrained neural network (NN) evolutionary search method to optimize AutoFuzz. Evaluation of our prototype on one state-of-the-art learning-based controller, two rule-based controllers, and one industrial-grade controller in five scenarios shows that AutoFuzz efficiently finds hundreds of traffic violations in high-fidelity simulation environments. For each scenario, AutoFuzz can find on average 10-39% more unique traffic violations than the best-performing baseline method. Further, fine-tuning the learning-based controller with the traffic violations found by AutoFuzz successfully reduced the traffic violations found in the new version of the AV controller software.