pith. machine review for the scientific record. sign in

arxiv: 2604.02647 · v1 · submitted 2026-04-03 · 💻 cs.SE

Recognition: no theorem link

Runtime Execution Traces Guided Automated Program Repair with Multi-Agent Debate

Jiaqing Wu , Tong Wu , Manqing Zhang , Yunwei Dong , Bo Shen
Authors on Pith no claims yet

Pith reviewed 2026-05-13 20:41 UTC · model grok-4.3

classification 💻 cs.SE
keywords automated program repairmulti-agent systemsruntime execution traceslarge language modelsDefects4Jpatch validationsoftware debugging
0
0 comments X

The pith

A multi-agent system treats runtime execution traces as objective constraints to correctly repair 392 defects on the Defects4J benchmark.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper claims that static LLM-based repair methods fail on complex logic errors because they lack dynamic runtime evidence and tend to overfit to test-passing patches. TraceRepair instead captures execution snapshots of critical variables via a probe agent and uses these facts as shared constraints for a committee of agents to cross-verify and refine candidate patches. This setup exposes logical inconsistencies that single-model approaches miss, leading to substantially more correct fixes than prior LLM methods. A sympathetic reader would care because the approach turns runtime data from optional input into a binding check that improves reliability of automated debugging.

Core claim

TraceRepair deploys a probe agent to record execution snapshots of selected variables during test execution, establishing an objective repair basis from actual state transitions. A committee of specialized agents then debates candidate patches against these runtime facts, iteratively exposing inconsistencies and refining proposals until they align with observed behavior, which yields 392 correctly fixed defects on Defects4J and demonstrates gains on a new set of recent bugs that arise from dynamic reasoning rather than memorization.

What carries the argument

The multi-agent debate framework in which runtime execution traces serve as shared objective constraints for patch validation, with a probe agent capturing snapshots and a committee performing cross-verification and refinement.

If this is right

  • Runtime traces used as constraints reduce reliance on coincidental test passage and improve logical correctness of generated patches.
  • Multi-agent cross-verification exposes inconsistencies that isolated LLM reasoning overlooks during patch generation.
  • Performance gains on both Defects4J and recent bugs stem from dynamic reasoning over the captured execution facts rather than static code patterns.
  • The framework generalizes beyond the training distribution because the traces supply fresh, program-specific evidence at repair time.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Extending the probe to capture additional dataflow or exception paths could further tighten the constraint set and reduce remaining false positives.
  • The same trace-constraint pattern might apply to other LLM tasks such as test generation or specification inference where behavioral evidence is available.
  • Selecting which variables to probe automatically, rather than manually, would be needed for broader industrial adoption.

Load-bearing premise

Runtime execution traces captured by the probe agent provide objective constraints that are sufficient to prevent overfitting to test-passing patches, and multi-agent cross-verification can reliably expose logical inconsistencies in candidate patches.

What would settle it

A collection of patches that satisfy all tests yet contradict the recorded runtime snapshots on the probed critical variables, or cases where the committee accepts incorrect patches without detecting their mismatch with observed execution states.

Figures

Figures reproduced from arXiv: 2604.02647 by Bo Shen, Jiaqing Wu, Manqing Zhang, Tong Wu, Yunwei Dong.

Figure 1
Figure 1. Figure 1: Compress-26 repair comparison between the static-only approach (left) and [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The Architecture of TraceRepair. The framework comprises a Trace-Guided Debate System interacting with a Universal Execution Sandbox. Limitations of Static Diagnosis. Without runtime context, the static agent cannot determine which part of the loop actually executed. As shown in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overlap analysis of correct fixes between [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Cumulative correct fixes across stages in [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
read the original abstract

Automated Program Repair (APR) struggles with complex logic errors and silent failures. Current LLM-based APR methods are mostly static, relying on source code and basic test outputs, which fail to accurately capture complex runtime behaviors and dynamic data dependencies. While incorporating runtime evidence like execution traces exposes concrete state transitions, a single LLM interpreting this in isolation often overfits to specific hypotheses, producing patches that satisfy tests by coincidence rather than correct logic. Therefore, runtime evidence should act as objective constraints rather than mere additional input. We propose TraceRepair, a multi-agent framework that leverages runtime facts as shared constraints for patch validation. A probe agent captures execution snapshots of critical variables to form an objective repair basis. Meanwhile, a committee of specialized agents cross-verifies candidate patches to expose inconsistencies and iteratively refine them. Evaluated on the Defects4J benchmark, TraceRepair correctly fixes 392 defects, substantially outperforming existing LLM-based approaches. Extensive experiments demonstrate improved efficiency and strong generalization on a newly constructed dataset of recent bugs, confirming that performance gains arise from dynamic reasoning rather than memorization.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims to introduce TraceRepair, a multi-agent framework for automated program repair that uses runtime execution traces captured by a probe agent to form objective constraints for patch validation. A committee of specialized agents cross-verifies candidate patches to expose inconsistencies and iteratively refines them. Evaluated on the Defects4J benchmark, it reports correctly fixing 392 defects and substantially outperforming existing LLM-based approaches, with additional experiments demonstrating improved efficiency and generalization on a newly constructed dataset of recent bugs.

Significance. If the empirical results hold under detailed scrutiny, this work could meaningfully advance LLM-based automated program repair by treating runtime traces as shared objective constraints rather than supplementary input, thereby reducing coincidental test-passing patches. The multi-agent cross-verification mechanism offers a structured way to detect logical inconsistencies, which addresses a known weakness in single-LLM repair methods. The evaluation on both Defects4J and a new recent-bug dataset provides a basis for assessing generalization beyond memorization.

major comments (2)
  1. [Abstract] Abstract: The central claim of 392 correct fixes on Defects4J is presented without accompanying details on the total number of bugs attempted, per-project breakdown, success rate, specific baseline methods and their fix counts, statistical significance tests, or ablation results on the probe-agent and committee components; this prevents full assessment of whether the reported improvement is attributable to the runtime-trace constraints and multi-agent debate.
  2. [Evaluation] Evaluation section: The assumption that probe-agent snapshots provide sufficient objective constraints is load-bearing for the claim of reduced overfitting, yet snapshots are captured only on the provided test executions; patches that match observed variable states on tested paths while altering unexercised control flow can still pass both the test suite and trace checks without restoring correct logic, and no additional validation (e.g., manual semantic analysis of fixed patches or extended test suites) is described to address this risk.
minor comments (2)
  1. Clarify the exact criteria used by the probe agent to select 'critical variables' for snapshotting and how the multi-agent debate protocol is implemented (e.g., number of rounds, voting mechanism).
  2. [Evaluation] Include a table in the evaluation section that directly compares fix counts, precision, and recall against the specific LLM-based baselines referenced in the related-work discussion.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment point by point below and indicate where revisions have been made to the next version of the paper.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of 392 correct fixes on Defects4J is presented without accompanying details on the total number of bugs attempted, per-project breakdown, success rate, specific baseline methods and their fix counts, statistical significance tests, or ablation results on the probe-agent and committee components; this prevents full assessment of whether the reported improvement is attributable to the runtime-trace constraints and multi-agent debate.

    Authors: The abstract serves as a concise summary, while the Evaluation section (Section 4) and associated tables provide the requested details: total bugs attempted (835 in Defects4J), per-project breakdown in Table 2, success rate (392/835), specific baselines with fix counts (e.g., comparisons to ChatRepair and others), statistical significance tests, and ablation results on the probe-agent and committee components in Section 4.3. To improve accessibility, we have revised the abstract to briefly note the total bugs attempted, overall success rate, and main baselines outperformed. We maintain that the full attribution of gains to runtime-trace constraints and multi-agent debate is best assessed from the detailed experiments rather than the abstract alone. revision: yes

  2. Referee: [Evaluation] Evaluation section: The assumption that probe-agent snapshots provide sufficient objective constraints is load-bearing for the claim of reduced overfitting, yet snapshots are captured only on the provided test executions; patches that match observed variable states on tested paths while altering unexercised control flow can still pass both the test suite and trace checks without restoring correct logic, and no additional validation (e.g., manual semantic analysis of fixed patches or extended test suites) is described to address this risk.

    Authors: This concern is valid and applies broadly to test-driven APR. Our design uses runtime snapshots as shared objective constraints across agents, with the committee performing cross-verification to surface logical inconsistencies that go beyond simple state matching on tested paths. However, we acknowledge that the original manuscript did not include manual semantic analysis or extended test suites. In the revised manuscript, we have added a dedicated paragraph in the Evaluation section discussing this limitation and reporting results from a manual review of 30 randomly sampled fixed patches confirming semantic correctness in the majority of cases. We have also executed the patches against any additional available tests in the benchmark for further validation. revision: partial

Circularity Check

0 steps flagged

No circularity in TraceRepair derivation or evaluation

full rationale

The paper proposes a multi-agent APR framework using probe-agent runtime snapshots as constraints and committee cross-verification for refinement. The headline result (392 correct fixes on Defects4J) is an empirical count from external benchmark evaluation, not a quantity derived from any internal equation, fitted parameter, or self-citation chain. No self-definitional steps, no predictions that reduce to fitted inputs, and no load-bearing self-citations or uniqueness theorems appear in the provided text. The method is self-contained against the Defects4J benchmark and a new external dataset.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The framework rests on the domain assumption that runtime snapshots form objective repair constraints and that agent debate improves logical correctness beyond test satisfaction.

free parameters (1)
  • committee size and agent specialization
    Number and roles of specialized agents are design choices whose values are not specified in the abstract.
axioms (1)
  • domain assumption Runtime execution traces provide objective constraints on patch correctness independent of test outcomes
    Invoked in the abstract as the basis for treating traces as shared validation facts rather than mere inputs.

pith-pipeline@v0.9.0 · 5487 in / 1142 out tokens · 39719 ms · 2026-05-13T20:41:27.438818+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. AuditRepairBench: A Paired-Execution Trace Corpus for Evaluator-Channel Ranking Instability in Agent Repair

    cs.AI 2026-05 unverdicted novelty 6.0

    AuditRepairBench supplies a large trace corpus and four screening methods that reduce evaluator-channel ranking instability in agent repair leaderboards by a mean of 62%.

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages · cited by 1 Pith paper · 4 internal anchors

  1. [1]

    Afsoon Afzal, Manish Motwani, Kathryn T Stolee, Yuriy Brun, and Claire Le Goues. 2019. SOSRepair: Expressive semantic search for real-world program repair.IEEE Transactions on Software Engineering (TSE)47, 10 (2019), 2162–2181

    work page 2019

  2. [2]

    Islem Bouzenia, Premkumar Devanbu, and Michael Pradel. 2025. RepairAgent: An Autonomous, LLM-Based Agent for Program Repair. InProceedings of the International Conference on Software Engineering (ICSE). IEEE Computer Society, 694–694

    work page 2025

  3. [3]

    Chi-Min Chan, Weize Chen, Yusheng Su, Jianxuan Yu, Wei Xue, Shanghang Zhang, Jie Fu, and Zhiyuan Liu. 2024. ChatEval: Towards Better LLM-based Eval- uators through Multi-Agent Debate. InProceedings of the International Conference on Learning Representations (ICLR)

    work page 2024

  4. [4]

    Mark Chen. 2021. Evaluating large language models trained on code.arXiv preprint arXiv:2107.03374(2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  5. [5]

    Xinyun Chen, Maxwell Lin, Nathanael Schärli, and Denny Zhou. 2024. Teach- ing Large Language Models to Self-Debug. InProceedings of the International Conference on Learning Representations (ICLR)

    work page 2024

  6. [6]

    Yilun Du, Shuang Li, Antonio Torralba, Joshua B Tenenbaum, and Igor Mordatch

    work page

  7. [7]

    InProceedings of the International Conference on Machine Learning (ICML)

    Improving factuality and reasoning in language models through multiagent debate. InProceedings of the International Conference on Machine Learning (ICML)

    work page

  8. [8]

    Xiang Gao, Bo Wang, Gregory J Duck, Ruyi Ji, Yingfei Xiong, and Abhik Roy- choudhury. 2021. Beyond tests: Program vulnerability repair via crash constraint extraction.ACM Transactions on Software Engineering and Methodology (TOSEM) 30, 2 (2021), 1–27

    work page 2021

  9. [9]

    Ali Ghanbari and Andrian Marcus. 2022. Patch correctness assessment in auto- mated program repair based on the impact of patches on production and test code. InProceedings of the ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA). 654–665

    work page 2022

  10. [10]

    Ali Ghanbari and Lingming Zhang. 2019. PraPR: Practical program repair via bytecode mutation. InProceedings of the IEEE/ACM International Conference on Automated Software Engineering (ASE). IEEE, 1118–1121

    work page 2019

  11. [11]

    Daya Guo, Qihao Zhu, Dejian Yang, Zhenda Xie, Kai Dong, Wentao Zhang, Guanting Chen, Xiao Bi, Yu Wu, YK Li, et al. 2024. DeepSeek-Coder: When the Large Language Model Meets Programming–The Rise of Code Intelligence.arXiv preprint arXiv:2401.14196(2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  12. [12]

    Jinru Hua, Mengshi Zhang, Kaiyuan Wang, and Sarfraz Khurshid. 2018. Towards practical program repair with on-demand candidate generation. InProceedings of the International Conference on Software Engineering (ICSE). 12–23

    work page 2018

  13. [13]

    Jie Huang, Xinyun Chen, Swaroop Mishra, Huaixiu Steven Zheng, Adams Wei Yu, Jason Roberts, and Denny Zhou. 2024. Large Language Models Cannot Self- Correct Reasoning Yet. InProceedings of the International Conference on Learning Representations (ICLR)

    work page 2024

  14. [14]

    Ziwei Ji, Nayeon Lee, Rita Frieske, Tiezheng Yu, Dan Su, Yan Xu, Etsuko Ishii, Ye Jin Bang, Andrea Madotto, and Pascale Fung. 2023. Survey of hallucination in natural language generation.ACM Computing Surveys (CSUR)55, 12 (2023), 1–38

    work page 2023

  15. [15]

    Jiajun Jiang, Yingfei Xiong, Hongyu Zhang, Qing Gao, and Xiangqun Chen

    work page

  16. [16]

    In Proceedings of the ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA)

    Shaping program repair space with existing patches and similar code. In Proceedings of the ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA). 298–309

    work page

  17. [17]

    Nan Jiang, Thibaud Lutellier, and Lin Tan. 2021. Cure: Code-aware neural machine translation for automatic program repair. InProceedings of the International Conference on Software Engineering (ICSE). IEEE, 1161–1173

    work page 2021

  18. [18]

    Carlos E Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan. 2024. SWE-BENCH: CAN LANGUAGE MODELS RESOLVE REAL-WORLD GITHUB ISSUES?. InProceedings of the International Conference on Learning Representations (ICLR)

    work page 2024

  19. [19]

    René Just, Darioush Jalali, and Michael D Ernst. 2014. Defects4J: A database of ex- isting faults to enable controlled testing studies for Java programs. InProceedings of the ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA). 437–440

    work page 2014

  20. [20]

    Dongsun Kim, Jaechang Nam, Jaewoo Song, and Sunghun Kim. 2013. Automatic patch generation learned from human-written patches. InProceedings of the International Conference on Software Engineering (ICSE). IEEE, 802–811

    work page 2013

  21. [21]

    Anil Koyuncu, Kui Liu, Tegawendé F Bissyandé, Dongsun Kim, Jacques Klein, Martin Monperrus, and Yves Le Traon. 2020. Fixminer: Mining relevant fix patterns for automated program repair.Empirical Software Engineering (ESE)25, 3 (2020), 1980–2024

    work page 2020

  22. [22]

    Xuan-Bach D Le, Duc-Hiep Chu, David Lo, Claire Le Goues, and Willem Visser

    work page

  23. [23]

    InProceedings of the ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE)

    S3: syntax-and semantic-guided repair synthesis via programming by ex- amples. InProceedings of the ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE). 593–604

    work page

  24. [24]

    Claire Le Goues, ThanhVu Nguyen, Stephanie Forrest, and Westley Weimer. 2011. Genprog: A generic method for automatic software repair.IEEE Transactions on Software Engineering (TSE)38, 1 (2011), 54–72

    work page 2011

  25. [25]

    Fengjie Li, Jiajun Jiang, Jiajun Sun, and Hongyu Zhang. 2025. Hybrid automated program repair by combining large language models and program analysis.ACM Transactions on Software Engineering and Methodology (TOSEM)34, 7 (2025), 1–28

    work page 2025

  26. [26]

    Raymond Li, Loubna Ben Allal, Yangtian Zi, Niklas Muennighoff, Denis Kocetkov, Chenghao Mou, Marc Marone, Christopher Akiki, Jia Li, Jenny Chim, et al. 2023. Starcoder: may the source be with you!arXiv preprint arXiv:2305.06161(2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  27. [27]

    Kui Liu, Anil Koyuncu, Dongsun Kim, and Tegawendé F Bissyandé. 2019. TBar: Revisiting template-based automated program repair. InProceedings of the ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA). 31– 42

    work page 2019

  28. [28]

    Thibaud Lutellier, Hung Viet Pham, Lawrence Pang, Yitong Li, Moshi Wei, and Lin Tan. 2020. Coconut: combining context-aware neural translation models using ensemble for program repair. InProceedings of the ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA). 101–114

    work page 2020

  29. [29]

    Matias Martinez and Martin Monperrus. 2016. Astor: A program repair library for java. InProceedings of the ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA). 441–444

    work page 2016

  30. [30]

    Sergey Mechtaev, Jooyong Yi, and Abhik Roychoudhury. 2015. Directfix: Looking for simple program repairs. InProceedings of the International Conference on Software Engineering (ICSE), Vol. 1. IEEE, 448–458

    work page 2015

  31. [31]

    Sergey Mechtaev, Jooyong Yi, and Abhik Roychoudhury. 2016. Angelix: Scalable multiline program patch synthesis via symbolic analysis. InProceedings of the International Conference on Software Engineering (ICSE). 691–701

    work page 2016

  32. [32]

    Martin Monperrus. 2018. Automatic software repair: A bibliography.ACM Computing Surveys (CSUR)51, 1 (2018), 1–24

    work page 2018

  33. [33]

    Noor Nashid, Mifta Sintaha, and Ali Mesbah. 2023. Retrieval-based prompt selection for code-related few-shot learning. InProceedings of the International Conference on Software Engineering (ICSE). IEEE, 2450–2462

    work page 2023

  34. [34]

    Hoang Duong Thien Nguyen, Dawei Qi, Abhik Roychoudhury, and Satish Chan- dra. 2013. Semfix: Program repair via semantic analysis. InProceedings of the International Conference on Software Engineering (ICSE). IEEE, 772–781

    work page 2013

  35. [35]

    Yuhua Qi, Xiaoguang Mao, Yan Lei, Ziying Dai, and Chengsong Wang. 2014. The strength of random search on automated program repair. InProceedings of the International Conference on Software Engineering (ICSE). 254–265

    work page 2014

  36. [36]

    Zichao Qi, Fan Long, Sara Achour, and Martin Rinard. 2015. An analysis of patch plausibility and correctness for generate-and-validate patch generation systems. InProceedings of the ACM SIGSOFT International Symposium on Software Testing Conference acronym ’XX, June 03–05, 2018, Woodstock, NY Jiaqing Wu, Tong Wu, Manqing Zhang, Yunwei Dong and Bo Shen and...

    work page 2015

  37. [37]

    Baptiste Roziere, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiao- qing Ellen Tan, Yossi Adi, Jingyu Liu, Romain Sauvestre, Tal Remez, et al. 2023. Code llama: Open foundation models for code.arXiv preprint arXiv:2308.12950 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  38. [38]

    Ripon K Saha, Yingjun Lyu, Hiroaki Yoshida, and Mukul R Prasad. 2017. Elixir: Effective object-oriented program repair. InProceedings of the IEEE/ACM Interna- tional Conference on Automated Software Engineering (ASE). IEEE, 648–659

    work page 2017

  39. [39]

    Seemanta Saha et al. 2019. Harnessing evolution for multi-hunk program repair. InProceedings of the International Conference on Software Engineering (ICSE). IEEE, 13–24

    work page 2019

  40. [40]

    Edward K Smith, Earl T Barr, Claire Le Goues, and Yuriy Brun. 2015. Is the cure worse than the disease? overfitting in automated program repair. InProceedings of the ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE). 532–543

    work page 2015

  41. [41]

    Ting Su, Yichen Yan, Jue Wang, Jingling Sun, Yiheng Xiong, Geguang Pu, Ke Wang, and Zhendong Su. 2021. Fully automated functional fuzzing of Android apps for detecting non-crashing logic bugs.Proceedings of the ACM on Programming Languages (OOPSLA)5, OOPSLA (2021), 1–31

    work page 2021

  42. [42]

    Yuxiang Wei, Chunqiu Steven Xia, and Lingming Zhang. 2023. Copiloting the copilots: Fusing large language models with completion engines for automated program repair. InProceedings of the ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE). 172–184

    work page 2023

  43. [43]

    Ming Wen, Junjie Chen, Rongxin Wu, Dan Hao, and Shing-Chi Cheung. 2018. Context-aware patch generation for better automated program repair. InProceed- ings of the International Conference on Software Engineering (ICSE). 1–11

    work page 2018

  44. [44]

    Chunqiu Steven Xia, Yifeng Ding, and Lingming Zhang. 2023. The Plastic Surgery Hypothesis in the Era of Large Language Models. InProceedings of the IEEE/ACM International Conference on Automated Software Engineering (ASE). 522–534

    work page 2023

  45. [45]

    Chunqiu Steven Xia, Yuxiang Wei, and Lingming Zhang. 2023. Automated program repair in the era of large pre-trained language models. InProceedings of the International Conference on Software Engineering (ICSE). IEEE, 1482–1494

    work page 2023

  46. [46]

    Chunqiu Steven Xia and Lingming Zhang. 2022. Less training, more repairing please: revisiting automated program repair via zero-shot learning. InProceedings of the ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE). 959–971

    work page 2022

  47. [47]

    Chunqiu Steven Xia and Lingming Zhang. 2024. Automated program repair via conversation: Fixing 162 out of 337 bugs for $0.42 each using chatgpt. In Proceedings of the ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA). 819–831

    work page 2024

  48. [48]

    Qi Xin and Steven P Reiss. 2017. Leveraging syntax-related code for automated program repair. InProceedings of the IEEE/ACM International Conference on Auto- mated Software Engineering (ASE). IEEE, 660–670

    work page 2017

  49. [49]

    Yingfei Xiong, Jie Wang, Runfa Yan, Jiachen Zhang, Shi Han, Gang Huang, and Lu Zhang. 2017. Precise condition synthesis for program repair. InProceedings of the International Conference on Software Engineering (ICSE). IEEE, 416–426

    work page 2017

  50. [50]

    Junjielong Xu, Ying Fu, Shin Hwei Tan, and Pinjia He. 2025. Aligning the objective of llm-based program repair. InProceedings of the International Conference on Software Engineering (ICSE). IEEE, 2548–2560

    work page 2025

  51. [51]

    Jifeng Xuan, Matias Martinez, Favio Demarco, Maxime Clement, Sebastian Lame- las Marcote, Thomas Durieux, Daniel Le Berre, and Martin Monperrus. 2016. Nopol: Automatic repair of conditional statement bugs in java programs.IEEE Transactions on Software Engineering (TSE)43, 1 (2016), 34–55

    work page 2016

  52. [52]

    John Yang, Carlos E Jimenez, Alexander Wettig, Kilian Lieret, Shunyu Yao, Karthik Narasimhan, and Ofir Press. 2024. Swe-agent: Agent-computer interfaces en- able automated software engineering.dvances in Neural Information Processing Systems (NeurIPS)37 (2024), 50528–50652

    work page 2024

  53. [53]

    He Ye, Matias Martinez, and Martin Monperrus. 2022. Neural program repair with execution-based backpropagation. InProceedings of the International Conference on Software Engineering (ICSE). 1506–1518

    work page 2022

  54. [54]

    He Ye, Aidan ZH Yang, Chang Hu, Yanlin Wang, Tao Zhang, and Claire Le Goues

    work page

  55. [55]

    AdverIntent-Agent: Adversarial Reasoning for Repair Based on Inferred Program Intent.Proceedings of the ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA)2, ISSTA (2025), 1398–1420

    work page 2025

  56. [56]

    Xin Yin, Chao Ni, Shaohua Wang, Zhenhao Li, Limin Zeng, and Xiaohu Yang

    work page

  57. [57]

    InProceedings of the ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA)

    Thinkrepair: Self-directed automated program repair. InProceedings of the ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA). 1274–1286

    work page

  58. [58]

    Yuan Yuan and Wolfgang Banzhaf. 2018. Arja: Automated repair of java pro- grams via multi-objective genetic programming.IEEE Transactions on Software Engineering (TSE)46, 10 (2018), 1040–1067

    work page 2018

  59. [59]

    Jiayi Zhang, Kai Huang, Jian Zhang, Yang Liu, and Chunyang Chen. 2025. Repair Ingredients Are All You Need: Improving Large Language Model-Based Program Repair via Repair Ingredients Search.arXiv preprint arXiv:2506.23100(2025)

    work page arXiv 2025

  60. [60]

    Quanjun Zhang, Chunrong Fang, Tongke Zhang, Bowen Yu, Weisong Sun, and Zhenyu Chen. 2023. Gamma: Revisiting template-based automated program re- pair via mask prediction. InProceedings of the IEEE/ACM International Conference on Automated Software Engineering (ASE). IEEE, 535–547

    work page 2023

  61. [61]

    Qihao Zhu, Zeyu Sun, Yuan-an Xiao, Wenjie Zhang, Kang Yuan, Yingfei Xiong, and Lu Zhang. 2021. A syntax-guided edit decoder for neural program repair. InProceedings of the ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE). 341–353

    work page 2021