pith. sign in

arxiv: 2606.21926 · v1 · pith:FFJICLKCnew · submitted 2026-06-20 · 💻 cs.SE

A11YRepair: Bridging Web Accessibility Barriers via Knowledge-Enhanced Divide-and-Conquer Repair

Kai Huang , Ling Zhu , Jian Zhang , Xiaofei Xie , Chunyang Chen This is my paper

Pith reviewed 2026-06-26 11:54 UTC · model grok-4.3

classification 💻 cs.SE
keywords web accessibilityautomated program repairLLM-based repairdivide-and-conquerWCAGA11YBenchfault localizationpatch synthesis
0
0 comments X

The pith

A11YRepair repairs clusters of related web accessibility violations more effectively by grouping coordinated edits and adding WCAG knowledge to an LLM.

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

The paper argues that standard automated repair tools fail on real web accessibility problems because those problems usually appear as multiple linked violations across files rather than isolated bugs. A11YRepair addresses this by first clustering violations that need joint changes, then splitting each cluster by root cause so the LLM can focus on one issue at a time. It supplies the model with targeted WCAG rules during both localization and patch creation to keep fixes consistent and domain-aware. Experiments on a benchmark of 60 real projects show higher success rates and lower costs than baselines, and some of the generated patches were accepted into projects from major organizations.

Core claim

A11YRepair introduces a divide-and-conquer workflow that clusters violations requiring coordinated edits to reduce redundant localization, then decomposes each cluster by root cause so the LLM can generate focused and consistent patches. The framework further incorporates WCAG-driven knowledge to strengthen domain awareness during both fault localization and patch synthesis. On A11YBench of 60 real-world web projects, it achieves higher repair effectiveness and lower cost than state-of-the-art baselines, ablation studies confirm the importance of its design choices, and generated patches have been merged into open-source projects from Google, Microsoft, Facebook, IBM, K8s, Docker, and Alibab

What carries the argument

Divide-and-conquer workflow that clusters violations needing coordinated edits across files and decomposes them by root cause, augmented with selective WCAG knowledge for localization and synthesis.

If this is right

  • Existing single-fault APR systems can be extended to multi-fault web scenarios by adding an explicit clustering step before localization.
  • Selective injection of guideline documents like WCAG improves LLM patch quality without full retraining.
  • Repair cost drops when redundant localization across related violations is avoided.
  • Patches produced this way have a higher chance of acceptance in real open-source projects.
  • The same workflow structure applies to any domain where guideline documents exist and violations tend to co-occur.

Where Pith is reading between the lines

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

  • The clustering step could be reused for other multi-location repair tasks such as security or performance fixes that span several files.
  • If root-cause decomposition proves brittle on certain page structures, hybrid human-LLM review of clusters might be needed.
  • The approach suggests that guideline-aware repair could be tested on mobile or desktop accessibility issues beyond web WCAG.

Load-bearing premise

That violations can be reliably clustered by the need for coordinated edits and then decomposed by root cause such that an LLM supplied with WCAG knowledge will produce focused, consistent patches without introducing new violations.

What would settle it

A new set of web projects in which automatic clustering often groups unrelated violations or the resulting patches introduce additional WCAG violations at similar rates to baselines.

Figures

Figures reproduced from arXiv: 2606.21926 by Chunyang Chen, Jian Zhang, Kai Huang, Ling Zhu, Xiaofei Xie.

Figure 1
Figure 1. Figure 1: Multiple accessibility violations detected on the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: A motivating example of centrally repairable violations. Figure 3a shows three SVG elements that lack textual alternatives, pre￾venting screen reader users from understanding their navigation purposes. As shown in Figure 3b, the developer can resolve all three violations by ap￾plying the same repair behavior—adding descrip￾tive text to a shared com￾ponent file in a single re￾pair action. These viola￾tions … view at source ↗
Figure 4
Figure 4. Figure 4: A motivating example of introduced side effects after fixing. Figure 4a further shows that repairing the viola￾tions in Figure 3a individ￾ually introduces two vi￾olations ❶ ❷ caused by inconsistent names [31]. As shown in Figure 4b, this occurs when speech￾input users navigate by speaking the visible text, a point of failure occurs when the visible label and the accessible name are different. By analyzing … view at source ↗
Figure 2
Figure 2. Figure 2: The overview of A11YRepair. ❶-❷ It first groups all violations at the component and situation levels for fault localization and patch generation. ❸-❹ Then, it leverages chat and embedding models to locate buggy files and synthesize patch edits. ❺ Meanwhile, knowledge integration module selectively incorporates WCAG Requirements [29] and Techniques [57] for fault localization and patch generation. These obs… view at source ↗
Figure 5
Figure 5. Figure 5: The requirement is for targets to be at least 24 by 24 CSS pixels in size. In contrast, complex violations require pre￾cise interpretation of WCAG Success Cri￾teria. For instance, adjusting target size or spacing must sat￾isfy SC 2.5.8 Target Size [60], which explicitly requires pointer targets to be at least 24 by 24 CSS pixels. As shown in [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The area-based grouping strategy. follows the basic repair workflow, including fault localization and patch generation. Unlike prior work, A11YRepair introduces a hier￾archical planning mechanism that decomposes complex repair tasks into manageable subproblems and embeds accessibility-specific expertise to improve both repair effectiveness and efficiency. As shown in [PITH_FULL_IMAGE:figures/full_fig_p005… view at source ↗
Figure 8
Figure 8. Figure 8: The criterion/situation level grouping strategy. [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The fault localization process, including File Local [PITH_FULL_IMAGE:figures/full_fig_p006_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Example of Success Criteria and Techniques. [PITH_FULL_IMAGE:figures/full_fig_p007_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: A case of using LLM-based refining strategy. [PITH_FULL_IMAGE:figures/full_fig_p008_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: A case of embedding model localization. 3) Feature Retrieval. During fault localization, A11YRepair extracts attribute-level cues from each violation element (e.g., com￾ponent ID, key, and bounding box) to retrieve a focused set of struc￾turally related candidate files. The embedding model then performs semantic retrieval within this reduced search space, alleviating the limitations of relying solely on a… view at source ↗
read the original abstract

Web accessibility (A11Y), which ensures web content is perceivable and usable for users with disabilities, is a critical requirement for modern web applications. Yet existing tooling overwhelmingly focuses on detecting A11Y violations rather than repairing them. Automated program repair (APR) techniques appear promising for this setting, but our study shows that state-of-the-art APR systems perform poorly when applied to real-world A11Y violations. Unlike conventional sparse-bug scenarios, web A11Y issues often manifest as multiple structurally related violations per page, requiring coordinated edits across multiple files. Existing repair systems fail to manage this multi-fault scale, as they handle each bug individually without considering their relationships or incorporating domain rules such as the Web Content Accessibility Guidelines (WCAG). We propose A11YRepair, an LLM-based framework for web A11Y repair. A11YRepair introduces a divide-and-conquer workflow that first clusters violations requiring coordinated edits to reduce redundant localization, and then decomposes each cluster by root cause so the LLM can generate focused and consistent patches. The framework further incorporates WCAG-driven knowledge to strengthen domain awareness during both fault localization and patch synthesis. To support systematic evaluation, we construct A11YBench, a benchmark of 60 real-world web projects collected from GitHub. Experimental results show that A11YRepair achieves higher repair effectiveness and lower cost than state-of-the-art baselines, and ablation studies confirm the importance of its divide-and-conquer design and selective domain knowledge integration. Specifically, patches generated by A11YRepair have been merged into open-source projects from Google, Microsoft, Facebook, IBM, K8s, Docker, and Alibaba, demonstrating its practical value.

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 introduces A11YRepair, an LLM-based framework for automated repair of web accessibility (A11Y) violations. It proposes a divide-and-conquer workflow that clusters violations requiring coordinated edits across files, decomposes clusters by root cause, and augments both localization and patch generation with WCAG domain knowledge. The approach is evaluated on a newly constructed benchmark A11YBench of 60 real-world GitHub projects, claiming higher repair effectiveness and lower cost than state-of-the-art APR baselines, with ablation studies supporting the design choices and several generated patches merged into projects from major organizations.

Significance. If the experimental claims hold after verification of the evaluation protocol, the work addresses a genuine gap between A11Y detection tools and practical repair, where multi-fault, cross-file violations are common. The divide-and-conquer strategy plus selective WCAG integration is a plausible adaptation of APR ideas to this domain, and the reported upstream merges provide concrete evidence of practical utility beyond synthetic benchmarks.

major comments (2)
  1. [Evaluation / Experimental results] The central effectiveness claim rests on the divide-and-conquer workflow (clustering coordinated violations then root-cause decomposition) producing focused, consistent patches. However, the evaluation provides no quantitative metrics on clustering precision, recall, or inter-cluster consistency, leaving the key assumption that the LLM will generate non-conflicting edits untested.
  2. [Evaluation / Experimental results] No post-repair re-scanning of pages for newly introduced A11Y violations is reported. Without this check, it is impossible to confirm that the claimed effectiveness gains do not come at the cost of creating fresh violations, directly undermining the comparison to baselines.
minor comments (2)
  1. [Benchmark construction] The abstract states that A11YBench contains 60 projects but does not specify how projects or violations were selected or filtered; this detail belongs in the benchmark-construction subsection.
  2. [Ablation studies] Ablation results are summarized at a high level; tables should report per-metric deltas (e.g., success rate, cost) with statistical significance tests rather than qualitative statements.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the evaluation protocol. We address each major comment below and will incorporate revisions to strengthen the experimental claims.

read point-by-point responses

  1. Referee: [Evaluation / Experimental results] The central effectiveness claim rests on the divide-and-conquer workflow (clustering coordinated violations then root-cause decomposition) producing focused, consistent patches. However, the evaluation provides no quantitative metrics on clustering precision, recall, or inter-cluster consistency, leaving the key assumption that the LLM will generate non-conflicting edits untested.

    Authors: We acknowledge that the manuscript does not report direct quantitative metrics (precision, recall, or inter-cluster consistency) for the clustering step against any ground-truth clustering. The current evaluation instead relies on end-to-end repair success rates, ablation studies that isolate the contribution of clustering and decomposition, and the practical outcome of upstream patch merges. To directly test the assumption of non-conflicting edits, we will add a dedicated analysis of clustering quality in the revised manuscript, including precision/recall metrics computed against manually validated clusters on a subset of the benchmark. revision: yes

  2. Referee: [Evaluation / Experimental results] No post-repair re-scanning of pages for newly introduced A11Y violations is reported. Without this check, it is impossible to confirm that the claimed effectiveness gains do not come at the cost of creating fresh violations, directly undermining the comparison to baselines.

    Authors: We agree this verification is necessary to ensure the reported gains do not mask regressions. The original experiments did not include a systematic post-repair re-scan of all pages using the same detection tools. In the revision we will add this analysis, reporting the count of newly introduced violations (if any) for A11YRepair and each baseline, thereby confirming that effectiveness improvements are not achieved at the expense of additional violations. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper describes an LLM-based empirical framework for A11Y repair with divide-and-conquer clustering and WCAG integration, evaluated on a newly constructed A11YBench benchmark plus real-world patch merges. No equations, parameters, or predictions are present that could reduce to inputs by construction. No self-citation load-bearing steps, uniqueness theorems, or ansatzes appear in the derivation chain. The work is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no identifiable free parameters, axioms, or invented entities; the approach is presented as an empirical engineering framework rather than a derivation resting on new postulates.

pith-pipeline@v0.9.1-grok · 5848 in / 1298 out tokens · 35489 ms · 2026-06-26T11:54:23.618008+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

86 extracted references · 11 canonical work pages

  1. [1]

    2026.A11YBench

    A11YBench. 2026.A11YBench. Retrieved Mar 26, 2026 from https://sites.google. com/view/a11yrepair/a11ybench

    2026

  2. [2]

    2026.A11YRepair

    A11YRepair. 2026.A11YRepair. Retrieved Mar 26, 2026 from https://sites.google. com/view/a11yrepair

    2026

  3. [3]

    2026.Figshare

    A11YRepair. 2026.Figshare. Retrieved Mar 26, 2026 from https://doi.org/10.6084/ m9.figshare.31896229

    2026

  4. [4]

    2026.Impact

    A11YRepair. 2026.Impact. Retrieved Mar 26, 2026 from https://sites.google.com/ view/a11yrepair/impact

    2026

  5. [5]

    Suliman K Almasoud and Hassan I Mathkour. 2019. Instant adaptation enrich- ment technique to improve web accessibility for blind users. InProceedings of the 2019 3rd International Conference on Information System and Data Mining. 159–164

    2019

  6. [6]

    Ali S Alotaibi, Paul T Chiou, and William GJ Halfond. 2021. Automated repair of size-based inaccessibility issues in mobile applications. InProceedings of the 36th IEEE/ACM International Conference on Automated Software Engineering (ASE). 730–742

    2021

  7. [7]

    Abdulaziz Alshayban and Sam Malek. 2022. AccessiText: automated detection of text accessibility issues in Android apps. InProceedings of the 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE). 984–995

    2022

  8. [8]

    Abdul Haddi Amjad, Muhammad Danish, Bless Jah, and Muhammad Ali Gulzar

  9. [9]

    InProceedings of the 2025 IEEE/ACM 47th International Conference on Software Engineering (ICSE)

    Accessibility Issues in Ad-Driven Web Applications. InProceedings of the 2025 IEEE/ACM 47th International Conference on Software Engineering (ICSE). 2393–2405

    2025

  10. [10]

    Islem Bouzenia, Premkumar Devanbu, and Michael Pradel. 2025. RepairAgent: An Autonomous, LLM-Based Agent for Program Repair. InProceedings of the 2025 IEEE/ACM 47th International Conference on Software Engineering (ICSE). 2188–2200

    2025

  11. [11]

    Jieshan Chen, Chunyang Chen, Zhenchang Xing, Xiwei Xu, Liming Zhu, Guo- qiang Li, and Jinshui Wang. 2020. Unblind your apps: Predicting natural- language labels for mobile gui components by deep learning. InProceedings of the ACM/IEEE 42nd International Conference on Software Engineering (ICSE). 322–334

    2020

  12. [12]

    Zimin Chen, Yue Pan, Siyu Lu, Jiayi Xu, Claire Le Goues, Martin Monperrus, and He Ye. 2025. Prometheus: Unified Knowledge Graphs for Issue Resolution in Multilingual Codebases.arXiv preprint arXiv:2507.19942(2025)

    work page arXiv 2025

  13. [13]

    Paul T Chiou, Ali S Alotaibi, and William GJ Halfond. 2021. Detecting and localizing keyboard accessibility failures in web applications. InProceedings of the 29th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE). 855–867

    2021

  14. [14]

    Paul T Chiou, Robert Winn, Ali S Alotaibi, and William GJ Halfond. 2024. Au- tomatically Detecting Reflow Accessibility Issues in Responsive Web Pages. InProceedings of the 2024 IEEE/ACM 46th International Conference on Software Engineering (ICSE). 1811–1823

    2024

  15. [15]

    2026.Ant Design: An enterprise-class UI design language and React UI library

    Ant Design. 2026.Ant Design: An enterprise-class UI design language and React UI library. Retrieved Jan 13, 2026 from https://github.com/ant-design/ant-design.git

    2026

  16. [16]

    2026.Ant Design: Help designers/developers building beautiful products more flexible and working with happiness

    Ant Design. 2026.Ant Design: Help designers/developers building beautiful products more flexible and working with happiness. Retrieved Jan 13, 2026 from https: //ant.design/

    2026

  17. [17]

    Nadeen Fathallah, Daniel Hernández, and Steffen Staab. 2025. AccessGuru: Leveraging LLMs to Detect and Correct Web Accessibility Violations in HTML Code. InProceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility. 1–22

    2025

  18. [18]

    Mexhid Ferati and Lirim Sulejmani. 2016. Automatic Adaptation Techniques to Increase the Web Accessibility for Blind Users. InInternational 2016 - Posters’ Extended Abstracts - 18th International Conference, HCI, Vol. 618. 30–36

    2016

  19. [19]

    2025.Lighthouse

    Google. 2025.Lighthouse. Retrieved Oct 01, 2025 from https://developer.chrome. com/docs/lighthouse/

    2025

  20. [20]

    Jiahao Gu and Huaxun Huang. 2025. Characterizing and Repairing Color-Related Accessibility Issues in Android Apps. (2025), 1–13

    2025

  21. [21]

    Alexandra-Elena Guriţă and Radu-Daniel Vatavu. 2025. When LLM-Generated Code Perpetuates User Interface Accessibility Barriers, How Can We Break the Cycle. InProceedings of the 22nd International Web for All Conference (W4A’25)

    2025

  22. [22]

    Sabrina Haque and Christoph Csallner. 2025. Early Accessibility: Automating Alt-Text Generation for UI Icons During App Development.arXiv preprint arXiv:2504.13069(2025)

    work page arXiv 2025

  23. [23]

    I Tend to View Ads Almost Like a Pestilence

    Ziyao He, Syed Fatiul Huq, and Sam Malek. 2024. “I Tend to View Ads Almost Like a Pestilence”: On the Accessibility Implications of Mobile Ads for Blind Huang et al. Users. InProceedings of the 2024 IEEE/ACM 46th International Conference on Software Engineering (ICSE). 2432–2444

    2024

  24. [24]

    Ziyao He, Syed Fatiul Huq, and Sam Malek. 2025. Enhancing Web Accessibility: Automated Detection of Issues with Generative AI.Proceedings of the ACM on Software Engineering2, FSE (2025), 2264–2287

    2025

  25. [25]

    Calista Huang, Alyssa Ma, Suchir Vyasamudri, Eugenie Puype, Sayem Kamal, Juan Belza Garcia, Salar Cheema, and Michael Lutz. 2024. Access: Prompt engineering for automated web accessibility violation corrections.arXiv preprint arXiv:2401.16450(2024)

    work page arXiv 2024

  26. [26]

    Kai Huang, Zhengzi Xu, Su Yang, Hongyu Sun, Xuejun Li, Zheng Yan, and Yuqing Zhang. 2024. Evolving paradigms in automated program repair: Taxonomy, challenges, and opportunities.Comput. Surveys57, 2 (2024), 1–43

    2024

  27. [27]

    Kai Huang, Jian Zhang, Xiaofei Xie, and Chunyang Chen. 2025. Seeing is Fixing: Cross-Modal Reasoning with Multimodal LLMs for Visual Software Issue Repair. InProceedings of the 2025 40th IEEE/ACM International Conference on Automated Software Engineering (ASE). 1156–1168

    2025

  28. [28]

    Syed Fatiul Huq, Abdulaziz Alshayban, Ziyao He, and Sam Malek. 2023. #A11yDev: Understanding Contemporary Software Accessibility Practices from Twitter Conversations. InProceedings of the 2023 CHI Conference on Human Factors in Computing Systems. 1–18

    2023

  29. [29]

    IBM. 2025.IBM Accessibility Requirements: Here are the latest published rule sets in the IBM Equal Access Accessibility Checker.Retrieved Oct 01, 2025 from https://www.ibm.com/able/requirements/checker-rule-sets/

    2025

  30. [30]

    IBM. 2025.IBM Accessibility Requirements: This page lists the accessibility require- ments that need to be met for several standards and regulations.Retrieved Oct 01, 2025 from https://www.ibm.com/able/requirements/requirements/

    2025

  31. [31]

    2025.IBM Equal Access Toolkit

    IBM. 2025.IBM Equal Access Toolkit. Retrieved Oct 01, 2025 from https://www. ibm.com/able/toolkit/

    2025

  32. [32]

    2026.Label in Name

    IBM. 2026.Label in Name. Retrieved Jan 25, 2026 from https://www.ibm.com/ able/requirements/requirements/?version=v7_3#2_5_3

    2026

  33. [33]

    Shubhi Jain, Syed Fatiul Huq, Ziyao He, and Sam Malek. 2025. Automated Detection of Web Application Navigation Barriers for Screen Reader Users. In Proceedings of the 40th IEEE/ACM International Conference on Automated Software Engineering (ASE). 1906–1918

    2025

  34. [34]

    Carlos E Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik R Narasimhan. 2024. SWE-bench: Can Language Models Resolve Real-world Github Issues?. InThe Twelfth International Conference on Learning Representations

    2024

  35. [35]

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

    2014

  36. [36]

    Arun Krishnavajjala, SM Hasan Mansur, Justin Jose, and Kevin Moran. 2024. Motorease: Automated Detection of Motor Impairment Accessibility Issues in Mobile App UIs. InProceedings of the 2024 IEEE/ACM 46th International Conference on Software Engineering (ICSE). 2580–2592

    2024

  37. [37]

    Claire Le Goues, ThanhVu Nguyen, Stephanie Forrest, and Westley Weimer. 2012. GenProg: A Generic Method for Automatic Software Repair.IEEE Transactions on Software Engineering38, 1 (2012)

    2012

  38. [38]

    Cheryl Lee, Chunqiu Steven Xia, Longji Yang, Jen-tse Huang, Zhouruixing Zhu, Lingming Zhang, and Michael R Lyu. 2025. UniDebugger: Hierarchical Multi- Agent Framework for Unified Software Debugging. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP). 18248– 18277

    2025

  39. [39]

    Kui Liu, Anil Koyuncu, Dongsun Kim, and Tegawendé F Bissyandé. 2019. TBar: Revisiting template-based automated program repair. InProceedings of the 28th ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA). 31–42

    2019

  40. [40]

    Andrea Mangiatordi and Marco Lazzari. 2018. Combined use of artificial intelli- gence and crowdsourcing to provide alternative content for images on websites. InProceedings of the 2018 15th IEEE Annual Consumer Communications & Net- working Conference (CCNC). 1–6

    2018

  41. [41]

    Forough Mehralian, Titus Barik, Jeff Nichols, and Amanda Swearngin. 2024. Automated Code Fix Suggestions for Accessibility Issues in Mobile Apps.arXiv preprint arXiv:2408.03827(2024)

    work page arXiv 2024

  42. [42]

    Forough Mehralian, Ziyao He, and Sam Malek. 2025. Automated Accessibility Analysis of Dynamic Content Changes on Mobile Apps. InProceedings of the 2025 IEEE/ACM 47th International Conference on Software Engineering (ICSE). 2689–2701

    2025

  43. [43]

    Forough Mehralian, Navid Salehnamadi, Syed Fatiul Huq, and Sam Malek. 2022. Too much accessibility is harmful! automated detection and analysis of overly accessible elements in mobile apps. InProceedings of the 37th IEEE/ACM Interna- tional Conference on Automated Software Engineering (ASE). 1–13

    2022

  44. [44]

    Forough Mehralian, Navid Salehnamadi, and Sam Malek. 2021. Data-driven accessibility repair revisited: on the effectiveness of generating labels for icons in Android apps. InProceedings of the 29th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE). 107–118

    2021

  45. [45]

    Xiangxin Meng, Zexiong Ma, Pengfei Gao, and Chao Peng. 2024. An em- pirical study on llm-based agents for automated bug fixing.arXiv preprint arXiv:2411.10213(2024)

    work page arXiv 2024

  46. [46]

    Peya Mowar, Yi-Hao Peng, Jason Wu, Aaron Steinfeld, and Jeffrey P Bigham. 2025. CodeA11y: Making AI Coding Assistants Useful for Accessible Web Development. InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems. 1–15

    2025

  47. [47]

    Hoang Duong Thien Nguyen, Dawei Qi, Abhik Roychoudhury, and Satish Chan- dra. 2013. Semfix: Program repair via semantic analysis. InProceedings of the 2013 35th International Conference on Software Engineering (ICSE). 772–781

    2013

  48. [48]

    2025.o4-mini-2025-04-16

    OpenAI. 2025.o4-mini-2025-04-16. Retrieved Oct 01, 2025 from https://platform. openai.com/docs/models/o4-mini

    2025

  49. [49]

    2025.text-embedding-3-small

    OpenAI. 2025.text-embedding-3-small. Retrieved Oct 01, 2025 from https: //platform.openai.com/docs/models/text-embedding-3-small

    2025

  50. [50]

    Haifeng Ruan, Yuntong Zhang, and Abhik Roychoudhury. 2025. SpecRover: Code Intent Extraction via LLMs. InProceedings of the 2025 IEEE/ACM 47th International Conference on Software Engineering (ICSE). 963–974

    2025

  51. [51]

    Navid Salehnamadi, Forough Mehralian, and Sam Malek. 2022. Groundhog: An automated accessibility crawler for mobile apps. InProceedings of the 37th IEEE/ACM International Conference on Automated Software Engineering (ASE). 1–12

    2022

  52. [52]

    Letícia Seixas Pereira, João Guerreiro, André Rodrigues, Tiago Guerreiro, and Carlos Duarte. 2024. From Automation to User Empowerment: Investigating the Role of a Semi-automatic Tool in Social Media Accessibility.ACM Transactions on Accessible Computing17, 3 (2024), 1–25

    2024

  53. [53]

    Aditya Bharat Soni, Boxuan Li, Xingyao Wang, Valerie Chen, and Graham Neubig

  54. [54]

    arXiv preprint arXiv:2506.03011(2025)

    Coding Agents with Multimodal Browsing are Generalist Problem Solvers. arXiv preprint arXiv:2506.03011(2025)

    work page arXiv 2025

  55. [55]

    2025.Axe accessibility testing tools are the best on the planet

    Deque Systems. 2025.Axe accessibility testing tools are the best on the planet. Retrieved Oct 01, 2025 from https://www.deque.com/axe/

    2025

  56. [56]

    Mahan Tafreshipour, Anmol Deshpande, Forough Mehralian, Iftekhar Ahmed, and Sam Malek. 2024. Ma11y: A Mutation Framework for Web Accessibility Testing. InProceedings of the 33rd ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA). 100–111

    2024

  57. [57]

    SWE-Bench Team. 2025. SWE-Bench Multimodal Leaderboard. https://www. swebench.com/index.html#multimodal. Accessed: 2025-10-23

    2025

  58. [58]

    Trae Research Team, Pengfei Gao, Zhao Tian, Xiangxin Meng, Xinchen Wang, Ruida Hu, Yuanan Xiao, Yizhou Liu, Zhao Zhang, Junjie Chen, Cuiyun Gao, Yun Lin, Yingfei Xiong, Chao Peng, and Xia Liu. 2025. Trae Agent: An LLM-based Agent for Software Engineering with Test-time Scaling. (2025). arXiv:2507.23370

    work page arXiv 2025

  59. [59]

    2025.Techniques for WCAG 2.2

    WAI. 2025.Techniques for WCAG 2.2. Retrieved Dec 02, 2025 from https: //www.w3.org/WAI/WCAG22/Techniques/

    2025

  60. [60]

    2025.Understanding SC 1.1.1: Non-text Content (Level A)

    WAI. 2025.Understanding SC 1.1.1: Non-text Content (Level A). Retrieved Oct 01, 2025 from https://www.w3.org/WAI/WCAG22/Understanding/non-text- content.html

    2025

  61. [61]

    2025.Understanding SC 1.4.3: Contrast (Minimum) (Level AA)

    WAI. 2025.Understanding SC 1.4.3: Contrast (Minimum) (Level AA). Retrieved Oct 01, 2025 from https://www.w3.org/WAI/WCAG22/Understanding/contrast- minimum.html

    2025

  62. [62]

    2025.Understanding SC 2.5.8: Target Size (Minimum) (Level AA)

    WAI. 2025.Understanding SC 2.5.8: Target Size (Minimum) (Level AA). Retrieved Oct 01, 2025 from https://www.w3.org/WAI/WCAG22/Understanding/target- size-minimum

    2025

  63. [63]

    2025.Web Content Accessibility Guidelines (WCAG) 2.2

    WAI. 2025.Web Content Accessibility Guidelines (WCAG) 2.2. Retrieved Dec 02, 2025 from https://www.w3.org/TR/WCAG22/

    2025

  64. [64]

    Yuxuan Wan, Chaozheng Wang, Yi Dong, Wenxuan Wang, Shuqing Li, Yintong Huo, and Michael Lyu. 2025. Divide-and-Conquer: Generating UI Code from Screenshots.Proceedings of the ACM on Software Engineering2, FSE (2025), 2099–2122

    2025

  65. [65]

    Xu, Xiangru Tang, Mingchen Zhuge, Jiayi Pan, Yueqi Song, Bowen Li, Jaskirat Singh, Hoang H

    Xingyao Wang, Boxuan Li, Yufan Song, Frank F. Xu, Xiangru Tang, Mingchen Zhuge, Jiayi Pan, Yueqi Song, Bowen Li, Jaskirat Singh, Hoang H. Tran, Fuqiang Li, Ren Ma, Mingzhang Zheng, Bill Qian, Yanjun Shao, Niklas Muennighoff, Yizhe Zhang, Binyuan Hui, Junyang Lin, Robert Brennan, Hao Peng, Heng Ji, and Graham Neubig. 2025. OpenHands: An Open Platform for A...

    2025

  66. [66]

    2025.The 2025 report on the accessibility of the top 1,000,000 home pages

    WebAIM. 2025.The 2025 report on the accessibility of the top 1,000,000 home pages. Retrieved Oct 01, 2025 from https://webaim.org/projects/million/

    2025

  67. [67]

    2025.W A VE Web Accessibility Evaluation Tools

    WebAIM. 2025.W A VE Web Accessibility Evaluation Tools. Retrieved Oct 01, 2025 from https://wave.webaim.org/

    2025

  68. [68]

    Shaomei Wu, Jeffrey Wieland, Omid Farivar, and Julie Schiller. 2017. Automatic alt-text: Computer-generated image descriptions for blind users on a social net- work service. InProceedings of the 2017 ACM Conference on Computer Supported Cooperative Work and Social Computing. 1180–1192

    2017

  69. [69]

    Chunqiu Steven Xia, Yinlin Deng, Soren Dunn, and Lingming Zhang. 2025. Demystifying llm-based software engineering agents.Proceedings of the ACM on Software Engineering2, FSE (2025), 801–824

    2025

  70. [70]

    Chunqiu Steven Xia, Zhe Wang, Yan Yang, Yuxiang Wei, and Lingming Zhang

  71. [71]

    A11YRepair: Bridging Web Accessibility Barriers via Knowledge-Enhanced Divide-and-Conquer Repair

    Live-SWE-agent: Can Software Engineering Agents Self-Evolve on the Fly?arXiv preprint arXiv:2511.13646(2025). A11YRepair: Bridging Web Accessibility Barriers via Knowledge-Enhanced Divide-and-Conquer Repair

    work page arXiv 2025

  72. [72]

    Chunqiu Steven Xia and Lingming Zhang. 2022. Less training, more repairing please: revisiting automated program repair via zero-shot learning. InProceedings of the 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE). 959–971

    2022

  73. [73]

    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 33rd ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA). 819–831

    2024

  74. [74]

    Boyang Yang, Zijian Cai, Fengling Liu, Bach Le, Lingming Zhang, Tegawendé F Bissyandé, Yang Liu, and Haoye Tian. 2025. A Survey of LLM-based Automated Program Repair: Taxonomies, Design Paradigms, and Applications.arXiv preprint arXiv:2506.23749(2025)

    work page arXiv 2025

  75. [75]

    John Yang, Carlos E Jimenez, Alexander Wettig, Kilian Lieret, Shunyu Yao, Karthik Narasimhan, and Ofir Press. 2024. Swe-agent: Agent-computer interfaces enable automated software engineering.Advances in Neural Information Processing Systems37 (2024), 50528–50652

    2024

  76. [76]

    John Yang, Carlos E Jimenez, Alex L Zhang, Kilian Lieret, Joyce Yang, Xindi Wu, Ori Press, Niklas Muennighoff, Gabriel Synnaeve, Karthik R Narasimhan, Diyi Yang, Sida Wang, and Ofir Press. 2025. SWE-bench Multimodal: Do AI Systems Generalize to Visual Software Domains?. InThe Thirteenth International Conference on Learning Representations

    2025

  77. [77]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. 2023. React: Synergizing reasoning and acting in language models. InThe eleventh International Conference on Learning Representations. 1–33

    2023

  78. [78]

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

  79. [79]

    AdverIntent-Agent: Adversarial Reasoning for Repair Based on Inferred Program Intent.Proceedings of the ACM on Software Engineering2, ISSTA (2025), 1398–1420

    2025

  80. [80]

    Quanjun Zhang, Chunrong Fang, Yang Xie, YuXiang Ma, Weisong Sun, Yun Yang, and Zhenyu Chen. 2024. A systematic literature review on large language models for automated program repair.arXiv preprint arXiv:2405.01466(2024)

    work page arXiv 2024

Showing first 80 references.