---
title: "CodeScout — paper by Suri et al. (2026) — summary and ideas"
description: "It builds a repository graph, uses an LLM to scope the relevant parts of the codebase, and examines candidate code elements… 24 references."
authors: ["Manan Suri", "Xiangci Li", "Mehdi Shojaie", "Songyang Han", "Chao-Chun Hsu", "Shweta Garg", "Aniket Anand Deshmukh", "Varun Kumar"]
year: 2026
venue: "Findings of the Association for Computational Linguistics: ACL 2026"
source: "https://aclanthology.org/2026.findings-acl.2032/"
doi: "10.18653/v1/2026.findings-acl.2032"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/suri-2026-codescout-contextual-problem-statement-enhancement/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/suri-2026-codescout-contextual-problem-statement-enhancement/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/25/) / CodeScout: Contextual Problem Statement Enhancement for Software Agents

# CodeScout: Contextual Problem Statement Enhancement for Software Agents

**Authors:** [Manan Suri](https://md.research.elamur.ai/free/en/authors/manan-suri/), [Xiangci Li](https://md.research.elamur.ai/free/en/authors/xiangci-li/), Mehdi Shojaie, Songyang Han, Chao-Chun Hsu, Shweta Garg, Aniket Anand Deshmukh and [Varun Kumar](https://md.research.elamur.ai/free/en/authors/varun-kumar/) · **Year:** 2026 · **Venue:** [Findings of the Association for Computational Linguistics: ACL 2026](https://md.research.elamur.ai/free/en/venues/findings/) ([Findings of the Association for Computational Linguistics 2026](https://md.research.elamur.ai/free/en/venues/findings/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.findings-acl.2032/) · [PDF](https://aclanthology.org/2026.findings-acl.2032.pdf) · [DOI 10.18653/v1/2026.findings-acl.2032](https://doi.org/10.18653/v1/2026.findings-acl.2032) · [OpenAlex](https://openalex.org/W7166843774)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Coding agents often fail when they receive vague issue descriptions that omit reproduction steps, expected behavior or pointers into the code. Such failures tend to produce long trajectories in which the agent either explores without focus or keeps retrying the same patch. CodeScout tackles this by rewriting the issue before any agent starts work. It builds a repository graph, uses an LLM to scope the relevant parts of the codebase, and examines candidate code elements from several angles, covering possible fixes and areas worth exploring. It then produces a richer problem statement with reproduction steps, expected behavior, exploration hints and fix hints. The agent scaffold itself is not changed, so the method can sit in front of any agent. On SWEBench-Verified, tested with several models and scaffolds, the method raised resolution rates by about 20%, solving up to 27 more issues than the default setup. The work suggests that improving the input is a useful complement to building larger models or more complex agents.

## Key points
- Evaluated on SWEBench-Verified (Python only) with three LLMs (DeepSeek R1, GPT-5-mini, Qwen3-Coder) and three scaffolds (SWE-Agent, OpenHands, Mini-SWE-Agent).
- Headline result: roughly 20% higher resolution rate and up to 27 extra resolved issues compared with the baseline.
- In the case study django__django-11790, the original issue led to a failed 21-step run, while the enhanced statement led to a fix in 6 steps.
- A separate augmentation pipeline beat letting the agent augment the issue itself during its run. LLM-based scoping worked better than BM25 retrieval.
- The augmenting model and the runtime agent can be different models. A strong augmenter lifts weak agents most, while a weaker augmenter gives only small gains to a strong agent (GPT-5-mini went from 194 to 196 resolved issues with DeepSeek R1 as augmenter).

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [University of Maryland, College Park](https://md.research.elamur.ai/free/en/universities/university-of-maryland-college-park/) — University, College Park, United States · 132 works · 25,380 citations
- [Amazon](https://md.research.elamur.ai/free/en/companies/amazon/) — Company, Seattle, United States · 112 works · 40,565 citations

📚 Shelves (3): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🗄️ Information Systems](https://md.research.elamur.ai/free/en/shelves/computer-science/information-systems/) · 1,437 works · [💻 Software](https://md.research.elamur.ai/free/en/shelves/computer-science/software/) · 92 works

## Cite

Cite as: Manan Suri, Xiangci Li, Mehdi Shojaie, Songyang Han, Chao-Chun Hsu, Shweta Garg, Aniket Anand Deshmukh and Varun Kumar. 2026. CodeScout: Contextual Problem Statement Enhancement for Software Agents. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{suri2026codescout,
  title = {{CodeScout: Contextual Problem Statement Enhancement for Software Agents}},
  author = {Manan Suri, Xiangci Li, Mehdi Shojaie, Songyang Han, Chao-Chun Hsu, Shweta Garg, Aniket Anand Deshmukh and Varun Kumar},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.2032},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.2032}
}
```

- [← previous CodeGuard: Improving LLM Guardrails in CS Education](https://md.research.elamur.ai/free/en/papers/raihan-2026-codeguard-improving-llm-guardrails-cs/)
- [next → Codezone Research Group at AraSentEval Shared Task: Arabic Sentiment Swap beyond Negation Prepending, Benchmarking Multilingual T5 against Large Language Models on the MA'AKS Corpus](https://md.research.elamur.ai/free/en/papers/bichi-2026-codezone-research-group-arasenteval-shared/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/25/)

## References

24 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Toufique Ahmed, Kunal Suresh Pai, Premkumar Devanbu, and Earl T Barr. 2024. Automatic semantic augmentation of language model prompts (for code summarization). pages 1–13. [find](https://scholar.google.com/scholar?q=Toufique+Ahmed%2C+Kunal+Suresh+Pai%2C+Premkumar+Devanbu%2C+and+Earl+T+Barr%2E+2024%2E+Automatic+semantic+augmentation+of+language+model+prompts+%28for+code+summarization%29%2E+pages+1%E2%80%9313%2E)
2. Shraddha Barke, Michael B. James, and Nadia Polikarpova. 2023. Grounded copilot: How programmers interact with code-generating models. Proceedings of the ACM on Programming Languages, 7(OOPSLA1):85–111. [find](https://scholar.google.com/scholar?q=Shraddha+Barke%2C+Michael+B%2E+James%2C+and+Nadia+Polikarpova%2E+2023%2E+Grounded+copilot%3A+How+programmers+interact+with+code%2Dgenerating+models%2E+Proceedings+of+the+ACM+on+Programming+Languages%2C+7%28OOPSLA1%29%3A85%E2%80%9311)
3. Islem Bouzenia, Premkumar Devanbu, and Michael Pradel. 2025. Repairagent: An autonomous, llm-based agent for program repair. In Proceedings of the 47th International Conference on Software Engineering, ICSE ’25, pages 2188–2200. IEEE Computer Society. for effective code search. 2015 IEEE 22nd Intern… [find](https://scholar.google.com/scholar?q=Islem+Bouzenia%2C+Premkumar+Devanbu%2C+and+Michael+Pradel%2E+2025%2E+Repairagent%3A+An+autonomous%2C+llm%2Dbased+agent+for+program+repair%2E+In+Proceedings+of+the+47th+International+Conference+on+Software+Engineering)
4. Xiangxin Meng, Zexiong Ma, Pengfei Gao, and Chao Peng. 2024. An empirical study on llm-based agents for automated bug fixing. arXiv preprint arXiv:2411.10213. — [arXiv:2411.10213](https://arxiv.org/abs/2411.10213) [find](https://scholar.google.com/scholar?q=Xiangxin+Meng%2C+Zexiong+Ma%2C+Pengfei+Gao%2C+and+Chao+Peng%2E+2024%2E+An+empirical+study+on+llm%2Dbased+agents+for+automated+bug+fixing%2E+arXiv+preprint+arXiv%3A2411%2E10213%2E+%E2%80%94+%5BarXiv%3A2411%2E10213%5D%28https%3A%2F%2Farxiv%2Eorg%2Fab)
5. Daye Nam, Ahmed Omran, Ambar Murillo, Saksham Thakur, Abner Araujo, Marcel Blistein, Alexander Frömmgen, Vincent Hellendoorn, and Satish Chandra. 2025. Prompting llms for code editing: Struggles and remedies. arXiv preprint arXiv:2504.20196. — [arXiv:2504.20196](https://arxiv.org/abs/2504.20196) [find](https://scholar.google.com/scholar?q=Daye+Nam%2C+Ahmed+Omran%2C+Ambar+Murillo%2C+Saksham+Thakur%2C+Abner+Araujo%2C+Marcel+Blistein%2C+Alexander+Fr%C3%B6mmgen%2C+Vincent+Hellendoorn%2C+and+Satish+Chandra%2E+2025%2E+Prompting+llms+for+code+editing%3A+Struggles+and+r)
6. Liming Nie, He Jiang, Zhilei Ren, Zeyi Sun, and Xiaochen Li. 2016. Query expansion based on crowd knowledge for code search. IEEE Transactions on Services Computing, 9:771–783. [find](https://scholar.google.com/scholar?q=Liming+Nie%2C+He+Jiang%2C+Zhilei+Ren%2C+Zeyi+Sun%2C+and+Xiaochen+Li%2E+2016%2E+Query+expansion+based+on+crowd+knowledge+for+code+search%2E+IEEE+Transactions+on+Services+Computing%2C+9%3A771%E2%80%93783%2E)
7. Islem Bouzenia and Michael Pradel. 2025. Understanding software engineering agents: A study of thought-action-result trajectories. arXiv preprint arXiv:2506.18824. — [arXiv:2506.18824](https://arxiv.org/abs/2506.18824) [find](https://scholar.google.com/scholar?q=Islem+Bouzenia+and+Michael+Pradel%2E+2025%2E+Understanding+software+engineering+agents%3A+A+study+of+thought%2Daction%2Dresult+trajectories%2E+arXiv+preprint+arXiv%3A2506%2E18824%2E+%E2%80%94+%5BarXiv%3A2506%2E18824%5D%28https%3A%2F%2Farxiv%2Eo)
8. Stephen Robertson, Hugo Zaragoza, and 1 others. 2009. The probabilistic relevance framework: Bm25 and beyond. Foundations and Trends® in Information Retrieval, 3(4):333–389. [find](https://scholar.google.com/scholar?q=Stephen+Robertson%2C+Hugo+Zaragoza%2C+and+1+others%2E+2009%2E+The+probabilistic+relevance+framework%3A+Bm25+and+beyond%2E+Foundations+and+Trends%C2%AE+in+Information+Retrieval%2C+3%284%29%3A333%E2%80%93389%2E)
9. Nan Jiang, Kevin Liu, Thibaud Lutellier, and Lin Tan. 2023. Impact of code language models on automated program repair. In Proceedings of the 45th International Conference on Software Engineering, ICSE ’23, pages 1430–1442. IEEE Press. [find](https://scholar.google.com/scholar?q=Nan+Jiang%2C+Kevin+Liu%2C+Thibaud+Lutellier%2C+and+Lin+Tan%2E+2023%2E+Impact+of+code+language+models+on+automated+program+repair%2E+In+Proceedings+of+the+45th+International+Conference+on+Software+Engineering%2C+ICS)
10. Francesco Romeo, Luigi Arena, Francesco Blefari, Francesco Aurelio Pironti, Matteo Lupinacci, and Angelo Furfaro. 2025. Arpaccino: An agenticrag for policy as code compliance. Preprint, arXiv:2507.10584. — [arXiv:2507.10584](https://arxiv.org/abs/2507.10584) [find](https://scholar.google.com/scholar?q=Francesco+Romeo%2C+Luigi+Arena%2C+Francesco+Blefari%2C+Francesco+Aurelio+Pironti%2C+Matteo+Lupinacci%2C+and+Angelo+Furfaro%2E+2025%2E+Arpaccino%3A+An+agenticrag+for+policy+as+code+compliance%2E+Preprint%2C+arXiv%3A2507%2E105)
11. Carlos E. Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan. 2023. [Swe-bench: Can language models resolve real-world github issues?](https://md.research.elamur.ai/free/en/papers/jimenez-2024-swe-bench-can-language-models/) Preprint, arXiv:2310.06770.
12. Gabriel Ryan, Siddhartha Jain, Mingyue Shang, Shiqi Wang, Xiaofei Ma, Murali Krishna Ramanathan, and Baishakhi Ray. 2024. Code-aware prompting: A study of coverage-guided test generation in regression setting using llm. Proceedings of the ACM on Software Engineering, 1(FSE):951–971. [find](https://scholar.google.com/scholar?q=Gabriel+Ryan%2C+Siddhartha+Jain%2C+Mingyue+Shang%2C+Shiqi+Wang%2C+Xiaofei+Ma%2C+Murali+Krishna+Ramanathan%2C+and+Baishakhi+Ray%2E+2024%2E+Code%2Daware+prompting%3A+A+study+of+coverage%2Dguided+test+generation+in+regression)
13. Matthew Jin, Syed Shahriar, Michele Tufano, Xin Shi, Shuai Lu, Neel Sundaresan, and Alexey Svyatkovskiy. 2023. Inferfix: End-to-end program repair with llms. Preprint, arXiv:2303.07263. — [arXiv:2303.07263](https://arxiv.org/abs/2303.07263) [find](https://scholar.google.com/scholar?q=Matthew+Jin%2C+Syed+Shahriar%2C+Michele+Tufano%2C+Xin+Shi%2C+Shuai+Lu%2C+Neel+Sundaresan%2C+and+Alexey+Svyatkovskiy%2E+2023%2E+Inferfix%3A+End%2Dto%2Dend+program+repair+with+llms%2E+Preprint%2C+arXiv%3A2303%2E07263%2E+%E2%80%94+%5BarXiv%3A2303%2E)
14. Anand Kumar, Vishal Khare, Deepak Sharma, Satyam Kumar, Vijay Saini, Anshul Yadav, Sachendra Jain, Ankit Rana, Pratham Verma, Vaibhav Meena, and Avinash Edubilli. 2025. Intuition to evidence: Measuring ai’s true impact on developer productivity. Preprint, arXiv:2509.19708. — [arXiv:2509.19708](https://arxiv.org/abs/2509.19708) [find](https://scholar.google.com/scholar?q=Anand+Kumar%2C+Vishal+Khare%2C+Deepak+Sharma%2C+Satyam+Kumar%2C+Vijay+Saini%2C+Anshul+Yadav%2C+Sachendra+Jain%2C+Ankit+Rana%2C+Pratham+Verma%2C+Vaibhav+Meena%2C+and+Avinash+Edubilli%2E+2025%2E+Intuition+to+evidence%3A+Measurin)
15. Caroline Lemieux, Jeevana Priya Inala, Shuvendu K. Lahiri, and Siddhartha Sen. 2023. Codamosa: Escaping coverage plateaus in test generation with pre-trained large language models. In Proceedings of the 45th International Conference on Software Engineering, pages 919–931, Melbourne, Australia. IEEE/… [find](https://scholar.google.com/scholar?q=Caroline+Lemieux%2C+Jeevana+Priya+Inala%2C+Shuvendu+K%2E+Lahiri%2C+and+Siddhartha+Sen%2E+2023%2E+Codamosa%3A+Escaping+coverage+plateaus+in+test+generation+with+pre%2Dtrained+large+language+models%2E+In+Proceedings+of+t)
16. Junwei Liu, Kaixin Wang, Yixuan Chen, Xin Peng, Zhenpeng Chen, Lingming Zhang, and Yiling Lou. 2024. Large language model-based agents for software engineering: A survey. Preprint, arXiv:2409.02977. — [arXiv:2409.02977](https://arxiv.org/abs/2409.02977) [find](https://scholar.google.com/scholar?q=Junwei+Liu%2C+Kaixin+Wang%2C+Yixuan+Chen%2C+Xin+Peng%2C+Zhenpeng+Chen%2C+Lingming+Zhang%2C+and+Yiling+Lou%2E+2024%2E+Large+language+model%2Dbased+agents+for+software+engineering%3A+A+survey%2E+Preprint%2C+arXiv%3A2409%2E02977%2E+%E2%80%94)
17. Meili Lu, Xiaobing Sun, Shaowei Wang, David Lo, and Yucong Duan. 2015. Query expansion via wordnet 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 Sh… [find](https://scholar.google.com/scholar?q=Meili+Lu%2C+Xiaobing+Sun%2C+Shaowei+Wang%2C+David+Lo%2C+and+Yucong+Duan%2E+2015%2E+Query+expansion+via+wordnet+Xingyao+Wang%2C+Boxuan+Li%2C+Yufan+Song%2C+Frank+F%2E+Xu%2C+Xiangru+Tang%2C+Mingchen+Zhuge%2C+Jiayi+Pan%2C+Yueqi+Song)
18. Yanlin Wang, Lianghong Guo, Ensheng Shi, Wenqing Chen, Jiachi Chen, Wanjun Zhong, Menghan Wang, Hui Li, Hongyu Zhang, Ziyu Lyu, and Zibin Zheng. 2023. You augment me: Exploring chatgpt-based data augmentation for semantic code search. 2023 IEEE International Conference on Software Maintenance and Ev… [find](https://scholar.google.com/scholar?q=Yanlin+Wang%2C+Lianghong+Guo%2C+Ensheng+Shi%2C+Wenqing+Chen%2C+Jiachi+Chen%2C+Wanjun+Zhong%2C+Menghan+Wang%2C+Hui+Li%2C+Hongyu+Zhang%2C+Ziyu+Lyu%2C+and+Zibin+Zheng%2E+2023%2E+You+augment+me%3A+Exploring+chatgpt%2Dbased+data+augm)
19. Yaqi Wang and Haipei Xu. 2024. Srsa: A cost-efficient strategy-router search agent for real-world human-machine interactions. Preprint, arXiv:2411.14574. — [arXiv:2411.14574](https://arxiv.org/abs/2411.14574) [find](https://scholar.google.com/scholar?q=Yaqi+Wang+and+Haipei+Xu%2E+2024%2E+Srsa%3A+A+cost%2Defficient+strategy%2Drouter+search+agent+for+real%2Dworld+human%2Dmachine+interactions%2E+Preprint%2C+arXiv%3A2411%2E14574%2E+%E2%80%94+%5BarXiv%3A2411%2E14574%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F241)
20. 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](https://md.research.elamur.ai/free/en/papers/yang-2024-swe-agent-agent-computer-interfaces/). In Advances in Neural Information Processing Systems, volume 37.
21. Zhiqiang Yuan, Mingwei Liu, Shiji Ding, Kaixin Wang, Yixuan Chen, Xin Peng, and Yiling Lou. 2024. Evaluating and improving ChatGPT for unit test generation. Proceedings of the ACM on Software Engineering, 1(FSE):1703–1726. [find](https://scholar.google.com/scholar?q=Zhiqiang+Yuan%2C+Mingwei+Liu%2C+Shiji+Ding%2C+Kaixin+Wang%2C+Yixuan+Chen%2C+Xin+Peng%2C+and+Yiling+Lou%2E+2024%2E+Evaluating+and+improving+ChatGPT+for+unit+test+generation%2E+Proceedings+of+the+ACM+on+Software+Engineer)
22. Output: List of exploration targets T {(type, name, reasoning)} = Stage 1: Exploration Target Identification Yuntong Zhang, Haifeng Ruan, Zhiyu Fan, and Abhik Roychoudhury. 2024. Autocoderover: Autonomous program improvement. In Proceedings of the 33rd ACM SIGSOFT International Symposium on Software… [find](https://scholar.google.com/scholar?q=Output%3A+List+of+exploration+targets+T+%7B%28type%2C+name%2C+reasoning%29%7D+%3D+Stage+1%3A+Exploration+Target+Identification+Yuntong+Zhang%2C+Haifeng+Ruan%2C+Zhiyu+Fan%2C+and+Abhik+Roychoudhury%2E+2024%2E+Autocoderover%3A+Autono)
23. You are analyzing a software bug report to identify which files and code components to investigate. PROBLEM STATEMENT: {problem_statement} REPOSITORY STRUCTURE: {file_tree} Task: Identify the most promising files, classes, and functions to explore for understanding this issue. Focus on: Albert Ziegl… [find](https://scholar.google.com/scholar?q=You+are+analyzing+a+software+bug+report+to+identify+which+files+and+code+components+to+investigate%2E+PROBLEM+STATEMENT%3A+%7Bproblem%5Fstatement%7D+REPOSITORY+STRUCTURE%3A+%7Bfile%5Ftree%7D+Task%3A+Identify+the+most+pro)
24. Files/classes mentioned in the problem statement • Related components that might be affected • Common entry points and core functionality • Test files that might reveal expected behavior [find](https://scholar.google.com/scholar?q=Files%2Fclasses+mentioned+in+the+problem+statement+%E2%80%A2+Related+components+that+might+be+affected+%E2%80%A2+Common+entry+points+and+core+functionality+%E2%80%A2+Test+files+that+might+reveal+expected+behavior)

*Updated 1 November 2026*