Paper review 2026-07-30: 25 candidates

Reply with commands, one per line:
`/approve all` · `/approve 1,3-5` · `/reject 2` · `/edit 3 category=world_terminal tags=benchmark` (edit implies approve; `tags=-` clears tags). Valid category keys: see taxonomy.json.

_Auto-skipped 110 out-of-scope; 0 failed (retried next run)._

### 1. Kimi K3: Open Frontier Intelligence
_ Kimi Team, Tongtong Bai, Yifan Bai, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.24653) · 📥 inbox
**proposed: `foundation_models`** · tags: `model`
> Kimi K3 is a 2.8T-parameter MoE frontier model with native vision, 1M-token context, and RL post-training across general, agentic, and coding domains. It achieves frontier-level performance on long-horizon coding, agentic, reasoning, and vision benchmarks, released as open weights.
A flagship general-purpose frontier foundation model spanning code production and agentic action, matching the foundation_models definition.

### 2. Try Again, Don't Look Back: Blind Resampling Outperforms Self-Repair in Small Code Models
_Yuvraj Verma_ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26117)
**proposed: `studies`** · tags: `empirical`
> A placebo-controlled empirical study across model scales shows blind resampling outperforms self-repair (feeding back failed code and test output) in small code models, due to anchoring on prior failed attempts. It isolates the value of feedback from the value of a retry, finding execution feedback adds little over a content-free placebo.
An empirical/behavioral study analyzing a design component of code agents (self-repair) rather than proposing a task-solving method.

### 3. (Im)Paired Programming: Coding Agents Improve Productivity but Harm Understanding
_Nishant Balepur, Connor Baumler, Valerie Chen, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26375)
**proposed: `studies`** · tags: `empirical`
> A controlled study with 54 students compares coding agents versus chatbot-based coding, finding agents boost task completion but harm code comprehension, especially with low-effort interaction patterns. The paper derives design implications for coding agent developers to preserve user understanding.
An empirical/behavioral study of how developers use coding agents, not a method proposing an agent to perform a task, so it falls under studies.

### 4. Impossible to hide secret ...: Uncovering Security and Privacy Issues in LLM-native IDEs
_Mostafijur Rahman Akhond, Md Afif Al Mamun, Gias Uddin, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26390)
**proposed: `studies`** · tags: `empirical`
> An empirical study mining 1.1M Reddit posts to identify and taxonomize security and privacy issues developers report while using LLM-native IDEs like Cursor, Copilot, and Codex. Finds most issues stem from system-level design rather than the underlying LLMs, revealing developer mistrust.
An empirical study whose object is the LLM-native coding tools/agents themselves rather than a proposed method for a task, so it belongs to studies.

### 5. Unpacking Vibe Coding: Help-Seeking Processes in Student-AI Interactions While Programming
_Daiana Rinja, Eduardo Araujo Oliveira, Sonsoles López-Pernas, et al._ · arXiv 2026/04 · [paper](https://arxiv.org/abs/2604.27134)
**proposed: `studies`** · tags: `empirical`
> An empirical study of 'vibe coding' analyzing 19,418 student-AI interaction turns to characterize help-seeking behaviors, finding top performers use instrumental inquiry while low performers rely on executive delegation. It argues for pedagogically aligned AI design that discourages passive delegation.
Object of study is human-AI coding interaction practice itself, matching the studies/empirical leaf.

### 6. A First Look at Coding Agents' Compliance with AI Contribution Rules in Open-Source Communities
_Wenhao Yang, Runzhi He, Minghui Zhou_ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26819)
**proposed: `studies`** · tags: `empirical`
> Studies whether real-world coding agents actually follow open-source repositories' AI contribution rules (bans, disclosure, verification gates), finding agents rarely proactively check rules and never refuse to contribute despite bans.
Empirical/behavioral study of coding-agent practice, proposing no new agent or task method, so it belongs to studies.

### 7. Can AI agents conduct open-ended AI research? Early evidence from two case studies
_Peter Kirgis, Sayash Kapoor, Andrew Schwartz, et al._ · NeurIPS 2026 · [paper](https://arxiv.org/abs/2607.27191)
**proposed: `studies`** · tags: `empirical`
> Introduces 'shadow evaluations' where AI agents attempt the open-ended research question of unpublished papers, graded by the original authors. Frontier agents completed engineering work but failed at core research judgment, revealing five recurring failure modes.
An empirical/behavioral study analyzing agent research capability and failure modes, not a method proposing a task-performing agent.

### 8. REAP: Automatic Curation of Coding Agent Benchmarks from Interactive Production Usage
_Smriti Jha, Matteo Paltenghi, Chandra Maddila, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2604.01527)
**proposed: `software_debugging`** · tags: `benchmark`
> REAP is an automated pipeline that curates coding-agent benchmarks from real production developer-agent sessions, using LLM-based classification and stability checks to ensure trustworthy fail-to-pass evaluation tasks. It produces Harvest, a multi-language benchmark evaluating frontier models on real issue-resolution tasks.
Benchmark routes to software_debugging: tasks verify code changes against fail-to-pass tests derived from real developer issues, SWE-bench style.

### 9. MultiFixer: A Coordinator-Proposer Based Multi-Agent Framework For Fixing Multi-Hunk Bugs
_Haichuan Hu, Chunrong Fang, Ye Shang, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26591)
**proposed: `software_debugging`** · tags: none
> MultiFixer is a Coordinator-Proposer multi-agent framework for automated program repair targeting multi-hunk bugs, using tool-augmented analysis and iterative patch generation and refinement. It achieves state-of-the-art results on Defects4J and repairs real-world vulnerabilities on VUL4J, SEC-bench, and PatchEval.
An LLM agent framework that localizes and repairs reported bugs/patches, fitting the automated program repair definition of software_debugging.

### 10. Cross-Model Cross-Language AI Coding Agent Performance: Accuracy and Speed of Parallel CLRS Algorithms
_Shiqi Cheng, Evelyne Ringoot, Rabab Alomairy, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26083)
**proposed: `software_code_generation`** · tags: `empirical`
> An empirical cross-model, cross-language study comparing coding agents (Cursor Composer 2.0, GPT-5.4, Claude Sonnet 4.6) on generating parallel implementations of classic algorithms in C++, Python, and Julia. It measures correctness and speedup, finding performance is highly agent-, algorithm-, and language-dependent.
Empirical evaluation of coding agents on one activity (code generation) routes to that leaf per the single-activity study rule.

### 11. NISPO: Open-source IUPAC name generation tool
_Nicholas T. Runcie, Fergus Imrie, Charlotte M. Deane_ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26113)
**proposed: `software_code_generation`** · tags: none
> NISPO is an open-source RDKit-based Python package for IUPAC chemical name generation, built through an agentic self-improvement loop using OpenAI Codex (GPT-5.5) that iteratively coded and refined the tool against millions of molecules. The resulting package achieves 98.1% round-trip accuracy on held-out PubChem molecules.
An agent autonomously wrote and refined a software package from scratch, serving code generation.

### 12. TraceCoder: Explainable and Auditable Code Generation with Position-Key Snippet Versioning
_Rwaida Alssadi, Muntaser Syed, Balaji Kasula, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26307)
**proposed: `software_code_generation`** · tags: none
> TraceCoder is a code-generation approach that records provenance of each repair iteration (failure text, LLM explanation) via a position-key snippet versioning scheme and browser visualization, making agentic benchmark-driven code repair auditable. Evaluated on 30 algorithmic programming tasks with iterative LLM-driven repair loops.
The task is LLM-driven code generation refined via iterative test feedback on programming benchmarks, not resolution of a specific reported bug.

### 13. Better Call Grep: Evaluating and Improving Grep-Like Lexical Retrieval for Repository-Level Code Completion
_Baoyi Wang, Xingliang Wang, Guochang Li, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2601.23254)
**proposed: `software_code_generation`** · tags: none
> GrepRAG augments lightweight, index-free lexical retrieval (via LLM-generated ripgrep commands) with identifier-weighted re-ranking and structure-aware deduplication for repository-level code completion. It outperforms sophisticated graph-based RAG baselines on CrossCodeEval and RepoEval-Updated.
Serves repository-level code completion, a code-generation task, per master test step 2-3.

### 14. DHRCL:Training Code LLMs with Dense Hierarchical Rewards and Curriculum Learning
_Shuhang Wang, Ziming Li, Hui Cheng_ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26457)
**proposed: `software_code_generation`** · tags: none
> DHRCL is an RL training framework for code LLMs that decomposes reward into syntax, execution, unit-test pass rate, and structural similarity signals via a staged curriculum. It improves functional correctness of generated code across Qwen3 model scales.
A training method whose downstream task is producing correct code, serving code generation.

### 15. Fewer Clarifications, Better Code: Benchmarking Cross-Session Personalized Ambiguity Adaptation in Coding Assistants
_Zijian Xu, Wenshuo Zhang, Zisen Qin, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26611)
**proposed: `software_code_generation`** · tags: `benchmark`
> CAPA is a benchmark for personalized ambiguity adaptation in coding assistants, testing whether resolved history from prior sessions helps a new session's coding assistant resolve recurring user-specific ambiguity with less clarification. It evaluates 12 LLMs on executable success and turns-to-completion across 600 sessions.
Task is producing correct executable code from ambiguous specs, a code-generation benchmark.

### 16. MRCoder: An Efficient Context Selecting Approach for Repository-Level Code Generation
_Peiding Wang, Li Zhang, Fang Liu_ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26805)
**proposed: `software_code_generation`** · tags: none
> MRCoder is a Map-Reduce context-selection framework for repository-level code generation that uses draft-guided selection to cut redundant context, improving accuracy while reducing tokens and inference time.
Serves repository-level code generation, matching software_code_generation.

### 17. VITAL-RAG: Invariance Race for Context Allocation in Coding Agents
_Zijian Lu, Yonghua Lu, Mingcai Chen, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26937)
**proposed: `software_code_generation`** · tags: none
> VITAL-RAG is a retrieval-augmented context allocation method for coding agents that groups code fragments by canonical object to reduce redundancy while preserving task-relevant local semantics under token budgets. It improves Recall@4K and Pass@1 on RepoBench, RepoClassBench, and RepoExec.
Serves repository-level code completion/generation, routing it to software_code_generation.

### 18. MindForge: Teaching Small Language Models Whole-Life-Cycle Software Engineering via Source-Free Program Synthesis
_Yihao Chen, Shi Chang, Khaled Chawa, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.27146)
**proposed: `software_code_generation`** · tags: `training-data` `model`
> MindForge builds source-free training environments from CLI programs and curates program-synthesis trajectories to fine-tune small LLMs for building programs entirely from scratch. The resulting model substantially improves ProgramBench pass rates and transfers gains to repo generation, translation, bug fixing, and feature-implementation benchmarks.
Downstream task served is from-scratch program synthesis, matching software_code_generation's whole-repository generation scope.

### 19. SpecFirst: Behavioral Specification Elicitation as a First-Class Step in Agent-Based Program Synthesis from Scratch
_Yihao Chen, Shi Chang, Feng Lin, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.27167)
**proposed: `software_code_generation`** · tags: `benchmark`
> SpecFirst introduces a two-stage framework that elicits a behavioral specification from a binary and documentation before code synthesis, targeting from-scratch program construction on ProgramBench. It improves test pass rates and exploration coverage over single-loop baselines across multiple model families.
Task is from-scratch program construction from specification, the core case of software_code_generation.

### 20. An Empirical Study of Foundation Models for Variability-Induced Compilation Errors in Configurable C Code
_Rohit Gheyi, Lucas Albuquerque, Márcio Ribeiro, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2601.16755)
**proposed: `software_infrastructure`** · tags: `empirical`
> An empirical study evaluating foundation models (GPT-OSS-20B, Gemini, Codex) on detecting and repairing variability-induced compilation errors hidden under untested feature combinations in configurable C code.
Task is restoring buildability of code under untested configurations, matching software_infrastructure's build-failure repair scope; single-activity study, not a field-wide survey.

### 21. Graph Is the Verifier: Agentic Reinforcement Learning for Interprocedural Vulnerability Detection
_Yikun Li, Ting Zhang, Jiakun Liu, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26656)
**proposed: `software_security`** · tags: none
> VulAgentRL is an agentic RL framework for interprocedural vulnerability detection that queries a Code Property Graph for cross-function evidence and uses exact-match graph verification as a training reward. It outperforms state-of-the-art detectors including frontier models under leakage-free evaluation.
Agent gathers evidence and localizes vulnerabilities in code, matching software_security.

### 22. HoF-Bench: Rediscovering Real AI-Discovered CVEs Without Frontier Models
_Petr Simecek, Elnaz Babayeva, Jiri Balhar, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.27030)
**proposed: `software_security`** · tags: `benchmark`
> HoF-Bench is a benchmark of 95 real AI-discovered CVEs used to test whether minimal LLM-based vulnerability analyzers (without frontier models) can rediscover known vulnerabilities in mature open-source C/C++ projects. A detector-blinded judge grades findings on code path, root cause, condition, and impact.
The served task is vulnerability detection/localization, placing it in software_security with a benchmark tag.

### 23. Readability-Robust Code Summarization via Meta Curriculum Learning
_Wenhao Zeng, Yitian Chai, Hao Zhou, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2601.05485)
**proposed: `software_comprehension`** · tags: none
> Empirically shows LLMs like GPT-4o and DeepSeek-V3 degrade sharply on code summarization when code has poor readability, then proposes RoFTCodeSum, a curriculum-plus-meta-learning fine-tuning method to improve robustness against obfuscated code.
Serves code comprehension/summarization task, which is the software_comprehension leaf.

### 24. ExplainBench: Evaluating Code Explanations from Agents
_Zhiyuan Pan, Sungmin Kang, Imam Nur Bani Yusuf, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26451)
**proposed: `software_comprehension`** · tags: `benchmark`
> ExplainBench is a benchmark evaluating whether coding agents' natural-language explanations of patches accurately describe buggy code behavior and patch effects. It finds explanation quality diverges from SWE-bench rankings and proposes an audit agent to improve trustworthiness.
Serves the task of understanding and documenting agent code changes, i.e. code comprehension/explanation, not judging change quality per se.

### 25. How Developers Experience Debugging Unfamiliar Codebases with Code Tours Generated and Evaluated by Local LLMs
_Balfroid Martin, Albert Julien, Aliti Dzenatan, et al._ · arXiv 2026/07 · [paper](https://arxiv.org/abs/2607.26987)
**proposed: `software_comprehension`** · tags: `empirical`
> A user study of 26 developers debugging unfamiliar Java codebases using LLM-generated 'code tours' (onboarding documentation), examining trust calibration and preferences for tour style, and finding LLM-based tour evaluators are unreliable. The tours aid code comprehension to support debugging of unfamiliar repositories.
An empirical study of a single tool for code documentation/comprehension, not a field-wide survey, so it routes to software_comprehension with the empirical tag rather than studies.

<details><summary>machine payload (do not edit)</summary>

```json
[{"paper": {"id": "2607.24653", "title": "Kimi K3: Open Frontier Intelligence", "authors": [" Kimi Team", "Tongtong Bai", "Yifan Bai", "Yiping Bao", "M. C.", "Jianfeng Cai", "Xinyuan Cai", "Peizhou Cao", "Yuxuan Cao", "Ziwei Chai", "Y. Charles", "H. S. Che", "Guanduo Chen", "Guangyu Chen", "Guanzheng Chen", "Huarong Chen", "Jia Chen", "Jianlong Chen", "Jun Chen", "Kexin Chen", "Peng Chen", "Ruijue Chen", "Wentao Chen", "Xin Chen", "Yang Chen", "Yanru Chen", "Yifei Chen", "Yingjiang Chen", "Yuankun Chen", "Yujie Chen", "Yutian Chen", "Zhirong Chen", "Dazhi Cheng", "Yean Cheng", "Jialei Cui", "Jingbing Cui", "Anqi Dai", "Jiaqi Deng", "Hao Ding", "Rui Ding", "Shaofeng Ding", "Mengfan Dong", "Mengnan Dong", "Yuhao Dong", "Yuxin Dong", "Angang Du", "Chenzhuang Du", "Dikang Du", "Jusen Du", "Yulun Du", "Yu Fan", "Jing Feng", "Qiulin Feng", "Yichen Feng", "Kelin Fu", "Qiang Fu", "Fuxuan Gao", "Hongcheng Gao", "Jingyue Gao", "Tong Gao", "Weijia Gao", "Shangyi Geng", "Jie Gong", "Linhu Gong", "Shengao Gong", "Xiaochen Gong", "Qizheng Gu", "Yicheng Gu", "Shuhao Guan", "Haiqing Guo", "Shiqi Guo", "Xiang Guo", "Zhengyan Guo", "Beixi Hao", "Wenxin Hao", "Xiaoru Hao", "Dailan He", "Haotian He", "Lehan He", "Qi He", "Weiran He", "Xinran He", "Xinyi He", "Yibo He", "Yunjia He", "Chao Hong", "Tiange Hong", "Hao Hu", "Jiaxi Hu", "Ruikun Hu", "Weiming Hu", "Yangyang Hu", "Zhenxing Hu", "Liang Hua", "Jinbin Huang", "Ke Huang", "Ruiyuan Huang", "Siying Huang", "Weixiao Huang", "Yan Huang", "Zhengjie Huang", "Zhiqi Huang", "Yulong Hui", "Chaobo Jia", "Yutong Jiang", "Zhejun Jiang", "Zuoyou Jiang", "Wenyi Jin", "Xinyi Jin", "Yu Jing", "Huanjun Kong", "Guokun Lai", "Aidi Li", "Cheng Li", "Chengyuan Li", "Cong Li", "Fang Li", "Guanyu Li", "Haoyang Li", "Jia Li", "Junxiong Li", "Lei Li", "Letian Li", "Lincan Li", "Weihong Li", "Wentao Li", "Xintong Li", "Yang Li", "Yishen Li", "Yiwei Li", "Yuxiao Li", "Zhaowei Li", "Zhaoxi Li", "Zheming Li", "Zhengxiao Li", "Zhiyuan Li", "Jiawei Lin", "Xiaohan Lin", "Yibo Lin", "Zichao Lin", "Ziyan Lin", "Bill Liu", "Boxiao Liu", "Chuan Liu", "Liang Liu", "Shaowei Liu", "Shudong Liu", "Shuran Liu", "Tianwei Liu", "Weizhou Liu", "Yangyang Liu", "Yanming Liu", "Yibo Liu", "Yipeng Liu", "Zhengying Liu", "Zhiheng Liu", "Enzhe Lu", "Haoyu Lu", "Linqiang Lu", "Tingzhan Lu", "Zhiyuan Lu", "Aotian Luo", "G. Luo", "Junyu Luo", "Yifan Luo", "B. Lyu", "Wenzhou Lyu", "Shaoguang Mao", "Yuan Mei", "Xin Men", "Minqing Ni", "Yixuan Niu", "Siyuan Pan", "Shujun Peng", "Zhangyang Qi", "Ruoyu Qin", "ZeChao Qin", "Zeyu Qin", "Haiquan Qiu", "Jianxin Qiu", "Jiezhong Qiu", "Bowen Qu", "Yuhao Qu", "Zeyu Shang", "Youbo Shao", "Han Shen", "Jincheng Shi", "Juanfeng Shi", "Lidong Shi", "Shengyuan Shi", "Wingchun Siu", "Pengwei Song", "Xiaoxi Song", "Jianlin Su", "Yunfeng Su", "Zhaochen Su", "Lin Sui", "Jingsong Sun", "Junyao Sun", "Shaoning Sun", "Shuzhe Sun", "Tongyu Sun", "Yujun Sun", "Yunpeng Tai", "Chuning Tang", "Heyi Tang", "Sirui Tang", "Zecheng Tang", "Chaoran Tian", "Rongpeng Tian", "Yu Tian", "Wei Tu", "Chensi Wang", "Chuang Wang", "Chunjie Wang", "Dinglu Wang", "Feng Wang", "Hailong Wang", "Haiming Wang", "Hao Wang", "Hao Wang", "Huaqing Wang", "Hui Wang", "Jiayi Wang", "Jinglong Wang", "Jinhong Wang", "Jiuzheng Wang", "Linian Wang", "Shaobo Wang", "Shenzhi Wang", "Shuyi Wang", "Si Wang", "Siyuan Wang", "Tianfu Wang", "Wenjue Wang", "Xingran Wang", "Xinmei Wang", "Xinyuan Wang", "Xusheng Wang", "Yalin Wang", "Yangkun Wang", "Yao Wang", "Yaoyu Wang", "Yejie Wang", "Yiqin Wang", "Yucheng Wang", "Yuzhi Wang", "Zhaoji Wang", "Zhaowei Wang", "Zhengtao Wang", "Zhenhao Wang", "Zhongsheng Wang", "Zifan Wang", "Chu Wei", "Ming Wei", "Shouxin Wei", "Zichen Wen", "Fan Wu", "Haoning Wu", "Rucong Wu", "Wenhao Wu", "Xiaoxue Wu", "Yingcong Wu", "Yongqi Wu", "Yuxin Wu", "Zijian Wu", "Xinglang Xian", "Chenxuan Xiang", "Yuye Xiang", "Bocheng Xiao", "Chenjun Xiao", "Xin Xiao", "Jin Xie", "Xiaotong Xie", "Yifeng Xie", "Zhe Xie", "Bowei Xing", "Yiming Xiong", "Baosheng Xu", "Boyu Xu", "Jiale Xu", "Jianfan Xu", "Jing Xu", "Jinjing Xu", "L. H. Xu", "Qingtao Xu", "Shuyao Xu", "Suting Xu", "Tiantian Xu", "Tianxiang Xu", "Weixin Xu", "Xinran Xu", "Yangchuan Xu", "Ye Xu", "Yueni Xu", "Ziyao Xu", "Haonan Xue", "Junjie Yan", "Yaoyao Yan", "Fan Yang", "Guangyao Yang", "Hao Yang", "Junwei Yang", "Ruoyu Yang", "Wenjie Yang", "Xiaofei Yang", "Xinyu Yang", "Yi Yang", "Yiling Yang", "Ying Yang", "Yuchen Yang", "Zhen Yang", "Zhilin Yang", "Zian Yang", "Zuhao Yang", "Haotian Yao", "Dan Ye", "Haoran Ye", "Wenjie Ye", "Zhanbo Ye", "Bohong Yin", "Haoxiang Yin", "Xietong Yin", "Chengzhen Yu", "Haozhen Yu", "Longhui Yu", "Shengnan Yu", "Shuying Yu", "Tianxiang Yu", "Enming Yuan", "Mengjie Yuan", "Tongtian Yue", "Wei Yue", "Yang Yue", "Dunyuan Zha", "Haobing Zhan", "B. H. Zhang", "Dehao Zhang", "Fei Zhang", "Hao Zhang", "Haoyuan Zhang", "Huanyu Zhang", "Jiapei Zhang", "Jiaxuan Zhang", "Jin Zhang", "Kaiyi Zhang", "Miaozhen Zhang", "Puqi Zhang", "Qinglei Zhang", "Rong Zhang", "Rui Zhang", "Shaoshuai Zhang", "Shiyi Zhang", "Xiaobin Zhang", "Xiaoyun Zhang", "Y. Zhang", "Yangkun Zhang", "Ye Zhang", "Yichi Zhang", "Yikun Zhang", "Yizhi Zhang", "Yongting Zhang", "Yu Zhang", "Yutao Zhang", "Yutong Zhang", "Zheng Zhang", "Zijing Zhang", "Bin Zhao", "Chenguang Zhao", "Feifan Zhao", "Jinglun Zhao", "Jinxiang Zhao", "Shuai Zhao", "Wenshuo Zhao", "Xiangyu Zhao", "Xuanle Zhao", "Yikai Zhao", "Zijia Zhao", "Haozhi Zheng", "Huabin Zheng", "Ruihan Zheng", "Shaojie Zheng", "Tengyang Zheng", "Haofeng Zhong", "Lei Zhong", "Longguang Zhong", "M. Zhou", "Qiankang Zhou", "Runjie Zhou", "Ruozhang Zhou", "Xinyu Zhou", "Yiqiao Zhou", "Zaida Zhou", "Jinguo Zhu", "Liya Zhu", "Xinhao Zhu", "Yangjunfeng Zhu", "Yuxuan Zhu", "Zhen Zhu", "Chen Zhuang", "Weiyu Zhuang", "Xinxing Zu"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-27", "links": {"paper": "https://arxiv.org/abs/2607.24653", "github": "", "website": ""}}, "category": "foundation_models", "tags": ["model"], "summary": "Kimi K3 is a 2.8T-parameter MoE frontier model with native vision, 1M-token context, and RL post-training across general, agentic, and coding domains. It achieves frontier-level performance on long-horizon coding, agentic, reasoning, and vision benchmarks, released as open weights.", "reason": "A flagship general-purpose frontier foundation model spanning code production and agentic action, matching the foundation_models definition.", "source": "inbox"}, {"paper": {"id": "2607.26117", "title": "Try Again, Don't Look Back: Blind Resampling Outperforms Self-Repair in Small Code Models", "authors": ["Yuvraj Verma"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-28", "links": {"paper": "https://arxiv.org/abs/2607.26117", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "A placebo-controlled empirical study across model scales shows blind resampling outperforms self-repair (feeding back failed code and test output) in small code models, due to anchoring on prior failed attempts. It isolates the value of feedback from the value of a retry, finding execution feedback adds little over a content-free placebo.", "reason": "An empirical/behavioral study analyzing a design component of code agents (self-repair) rather than proposing a task-solving method.", "source": "crawl"}, {"paper": {"id": "2607.26375", "title": "(Im)Paired Programming: Coding Agents Improve Productivity but Harm Understanding", "authors": ["Nishant Balepur", "Connor Baumler", "Valerie Chen", "Eunsol Choi", "Rachel Rudinger", "Jordan Lee Boyd-Graber"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.26375", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "A controlled study with 54 students compares coding agents versus chatbot-based coding, finding agents boost task completion but harm code comprehension, especially with low-effort interaction patterns. The paper derives design implications for coding agent developers to preserve user understanding.", "reason": "An empirical/behavioral study of how developers use coding agents, not a method proposing an agent to perform a task, so it falls under studies.", "source": "crawl"}, {"paper": {"id": "2607.26390", "title": "Impossible to hide secret ...: Uncovering Security and Privacy Issues in LLM-native IDEs", "authors": ["Mostafijur Rahman Akhond", "Md Afif Al Mamun", "Gias Uddin", "Song Wang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.26390", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "An empirical study mining 1.1M Reddit posts to identify and taxonomize security and privacy issues developers report while using LLM-native IDEs like Cursor, Copilot, and Codex. Finds most issues stem from system-level design rather than the underlying LLMs, revealing developer mistrust.", "reason": "An empirical study whose object is the LLM-native coding tools/agents themselves rather than a proposed method for a task, so it belongs to studies.", "source": "crawl"}, {"paper": {"id": "2604.27134", "title": "Unpacking Vibe Coding: Help-Seeking Processes in Student-AI Interactions While Programming", "authors": ["Daiana Rinja", "Eduardo Araujo Oliveira", "Sonsoles López-Pernas", "Mohammed Saqr", "Marcus Specht", "Kamila Misiejuk"], "venue": "arXiv 2026/04", "category": "", "published": "2026-04-29", "links": {"paper": "https://arxiv.org/abs/2604.27134", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "An empirical study of 'vibe coding' analyzing 19,418 student-AI interaction turns to characterize help-seeking behaviors, finding top performers use instrumental inquiry while low performers rely on executive delegation. It argues for pedagogically aligned AI design that discourages passive delegation.", "reason": "Object of study is human-AI coding interaction practice itself, matching the studies/empirical leaf.", "source": "crawl"}, {"paper": {"id": "2607.26819", "title": "A First Look at Coding Agents' Compliance with AI Contribution Rules in Open-Source Communities", "authors": ["Wenhao Yang", "Runzhi He", "Minghui Zhou"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.26819", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "Studies whether real-world coding agents actually follow open-source repositories' AI contribution rules (bans, disclosure, verification gates), finding agents rarely proactively check rules and never refuse to contribute despite bans.", "reason": "Empirical/behavioral study of coding-agent practice, proposing no new agent or task method, so it belongs to studies.", "source": "crawl"}, {"paper": {"id": "2607.27191", "title": "Can AI agents conduct open-ended AI research? Early evidence from two case studies", "authors": ["Peter Kirgis", "Sayash Kapoor", "Andrew Schwartz", "Stephan Rabanser", "David Africa", "Konstantinos Voudouris", "Viet Nguyen", "Toby Pilditch", "Magda Dubois", "Harry Coppock", "Cozmin Ududec", "Nitya Nadgir", "Matilda Orona", "Tilman Bayer", "Derrick Chan-Sew", "Yue Ling", "Abhishek Shetty", "Helen Toner", "Gillian Hadfield", "Seth Lazar", "Steve Newman", "Shoshannah Tekofsky", "Rishi Bommasani", "Arvind Narayanan"], "venue": "NeurIPS 2026", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27191", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "Introduces 'shadow evaluations' where AI agents attempt the open-ended research question of unpublished papers, graded by the original authors. Frontier agents completed engineering work but failed at core research judgment, revealing five recurring failure modes.", "reason": "An empirical/behavioral study analyzing agent research capability and failure modes, not a method proposing a task-performing agent.", "source": "crawl"}, {"paper": {"id": "2604.01527", "title": "REAP: Automatic Curation of Coding Agent Benchmarks from Interactive Production Usage", "authors": ["Smriti Jha", "Matteo Paltenghi", "Chandra Maddila", "Vijayaraghavan Murali", "Shubham Ugare", "Satish Chandra"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-28", "links": {"paper": "https://arxiv.org/abs/2604.01527", "github": "", "website": ""}}, "category": "software_debugging", "tags": ["benchmark"], "summary": "REAP is an automated pipeline that curates coding-agent benchmarks from real production developer-agent sessions, using LLM-based classification and stability checks to ensure trustworthy fail-to-pass evaluation tasks. It produces Harvest, a multi-language benchmark evaluating frontier models on real issue-resolution tasks.", "reason": "Benchmark routes to software_debugging: tasks verify code changes against fail-to-pass tests derived from real developer issues, SWE-bench style.", "source": "crawl"}, {"paper": {"id": "2607.26591", "title": "MultiFixer: A Coordinator-Proposer Based Multi-Agent Framework For Fixing Multi-Hunk Bugs", "authors": ["Haichuan Hu", "Chunrong Fang", "Ye Shang", "Jiawei Liu", "Weifeng Sun", "Guoqing Xie", "Chenxing Zhong", "Quanjun Zhang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.26591", "github": "", "website": ""}}, "category": "software_debugging", "tags": [], "summary": "MultiFixer is a Coordinator-Proposer multi-agent framework for automated program repair targeting multi-hunk bugs, using tool-augmented analysis and iterative patch generation and refinement. It achieves state-of-the-art results on Defects4J and repairs real-world vulnerabilities on VUL4J, SEC-bench, and PatchEval.", "reason": "An LLM agent framework that localizes and repairs reported bugs/patches, fitting the automated program repair definition of software_debugging.", "source": "crawl"}, {"paper": {"id": "2607.26083", "title": "Cross-Model Cross-Language AI Coding Agent Performance: Accuracy and Speed of Parallel CLRS Algorithms", "authors": ["Shiqi Cheng", "Evelyne Ringoot", "Rabab Alomairy", "Alan Edelman"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-26", "links": {"paper": "https://arxiv.org/abs/2607.26083", "github": "", "website": ""}}, "category": "software_code_generation", "tags": ["empirical"], "summary": "An empirical cross-model, cross-language study comparing coding agents (Cursor Composer 2.0, GPT-5.4, Claude Sonnet 4.6) on generating parallel implementations of classic algorithms in C++, Python, and Julia. It measures correctness and speedup, finding performance is highly agent-, algorithm-, and language-dependent.", "reason": "Empirical evaluation of coding agents on one activity (code generation) routes to that leaf per the single-activity study rule.", "source": "crawl"}, {"paper": {"id": "2607.26113", "title": "NISPO: Open-source IUPAC name generation tool", "authors": ["Nicholas T. Runcie", "Fergus Imrie", "Charlotte M. Deane"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-28", "links": {"paper": "https://arxiv.org/abs/2607.26113", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "NISPO is an open-source RDKit-based Python package for IUPAC chemical name generation, built through an agentic self-improvement loop using OpenAI Codex (GPT-5.5) that iteratively coded and refined the tool against millions of molecules. The resulting package achieves 98.1% round-trip accuracy on held-out PubChem molecules.", "reason": "An agent autonomously wrote and refined a software package from scratch, serving code generation.", "source": "crawl"}, {"paper": {"id": "2607.26307", "title": "TraceCoder: Explainable and Auditable Code Generation with Position-Key Snippet Versioning", "authors": ["Rwaida Alssadi", "Muntaser Syed", "Balaji Kasula", "Lamine Deen", "Majed Alotaibi", "Mohammed Alghamdi", "Tyler Ton", "Ali Alqarni", "Marius Silaghi"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-28", "links": {"paper": "https://arxiv.org/abs/2607.26307", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "TraceCoder is a code-generation approach that records provenance of each repair iteration (failure text, LLM explanation) via a position-key snippet versioning scheme and browser visualization, making agentic benchmark-driven code repair auditable. Evaluated on 30 algorithmic programming tasks with iterative LLM-driven repair loops.", "reason": "The task is LLM-driven code generation refined via iterative test feedback on programming benchmarks, not resolution of a specific reported bug.", "source": "crawl"}, {"paper": {"id": "2601.23254", "title": "Better Call Grep: Evaluating and Improving Grep-Like Lexical Retrieval for Repository-Level Code Completion", "authors": ["Baoyi Wang", "Xingliang Wang", "Guochang Li", "Chen Zhi", "Junxiao Han", "Xinkui Zhao", "Nan Wang", "Shuiguang Deng", "Jianwei Yin"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2601.23254", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "GrepRAG augments lightweight, index-free lexical retrieval (via LLM-generated ripgrep commands) with identifier-weighted re-ranking and structure-aware deduplication for repository-level code completion. It outperforms sophisticated graph-based RAG baselines on CrossCodeEval and RepoEval-Updated.", "reason": "Serves repository-level code completion, a code-generation task, per master test step 2-3.", "source": "crawl"}, {"paper": {"id": "2607.26457", "title": "DHRCL:Training Code LLMs with Dense Hierarchical Rewards and Curriculum Learning", "authors": ["Shuhang Wang", "Ziming Li", "Hui Cheng"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.26457", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "DHRCL is an RL training framework for code LLMs that decomposes reward into syntax, execution, unit-test pass rate, and structural similarity signals via a staged curriculum. It improves functional correctness of generated code across Qwen3 model scales.", "reason": "A training method whose downstream task is producing correct code, serving code generation.", "source": "crawl"}, {"paper": {"id": "2607.26611", "title": "Fewer Clarifications, Better Code: Benchmarking Cross-Session Personalized Ambiguity Adaptation in Coding Assistants", "authors": ["Zijian Xu", "Wenshuo Zhang", "Zisen Qin", "Rui Sheng", "Yushi Sun", "Huamin Qu", "Chuhan Shi"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.26611", "github": "", "website": ""}}, "category": "software_code_generation", "tags": ["benchmark"], "summary": "CAPA is a benchmark for personalized ambiguity adaptation in coding assistants, testing whether resolved history from prior sessions helps a new session's coding assistant resolve recurring user-specific ambiguity with less clarification. It evaluates 12 LLMs on executable success and turns-to-completion across 600 sessions.", "reason": "Task is producing correct executable code from ambiguous specs, a code-generation benchmark.", "source": "crawl"}, {"paper": {"id": "2607.26805", "title": "MRCoder: An Efficient Context Selecting Approach for Repository-Level Code Generation", "authors": ["Peiding Wang", "Li Zhang", "Fang Liu"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.26805", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "MRCoder is a Map-Reduce context-selection framework for repository-level code generation that uses draft-guided selection to cut redundant context, improving accuracy while reducing tokens and inference time.", "reason": "Serves repository-level code generation, matching software_code_generation.", "source": "crawl"}, {"paper": {"id": "2607.26937", "title": "VITAL-RAG: Invariance Race for Context Allocation in Coding Agents", "authors": ["Zijian Lu", "Yonghua Lu", "Mingcai Chen", "Yiping Zuo", "Xin He", "Weijun Wang", "Weibei Fan"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.26937", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "VITAL-RAG is a retrieval-augmented context allocation method for coding agents that groups code fragments by canonical object to reduce redundancy while preserving task-relevant local semantics under token budgets. It improves Recall@4K and Pass@1 on RepoBench, RepoClassBench, and RepoExec.", "reason": "Serves repository-level code completion/generation, routing it to software_code_generation.", "source": "crawl"}, {"paper": {"id": "2607.27146", "title": "MindForge: Teaching Small Language Models Whole-Life-Cycle Software Engineering via Source-Free Program Synthesis", "authors": ["Yihao Chen", "Shi Chang", "Khaled Chawa", "Feng Lin", "Boyuan Chen", "Shaowei Wang", "Ahmed E. Hassan"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27146", "github": "", "website": ""}}, "category": "software_code_generation", "tags": ["training-data", "model"], "summary": "MindForge builds source-free training environments from CLI programs and curates program-synthesis trajectories to fine-tune small LLMs for building programs entirely from scratch. The resulting model substantially improves ProgramBench pass rates and transfers gains to repo generation, translation, bug fixing, and feature-implementation benchmarks.", "reason": "Downstream task served is from-scratch program synthesis, matching software_code_generation's whole-repository generation scope.", "source": "crawl"}, {"paper": {"id": "2607.27167", "title": "SpecFirst: Behavioral Specification Elicitation as a First-Class Step in Agent-Based Program Synthesis from Scratch", "authors": ["Yihao Chen", "Shi Chang", "Feng Lin", "Khaled Chawa", "Boyuan Chen", "Shaowei Wang", "Ahmed E. Hassan"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27167", "github": "", "website": ""}}, "category": "software_code_generation", "tags": ["benchmark"], "summary": "SpecFirst introduces a two-stage framework that elicits a behavioral specification from a binary and documentation before code synthesis, targeting from-scratch program construction on ProgramBench. It improves test pass rates and exploration coverage over single-loop baselines across multiple model families.", "reason": "Task is from-scratch program construction from specification, the core case of software_code_generation.", "source": "crawl"}, {"paper": {"id": "2601.16755", "title": "An Empirical Study of Foundation Models for Variability-Induced Compilation Errors in Configurable C Code", "authors": ["Rohit Gheyi", "Lucas Albuquerque", "Márcio Ribeiro", "Eduardo Almeida", "Danyllo Albuquerque", "Mirko Perkusich"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-28", "links": {"paper": "https://arxiv.org/abs/2601.16755", "github": "", "website": ""}}, "category": "software_infrastructure", "tags": ["empirical"], "summary": "An empirical study evaluating foundation models (GPT-OSS-20B, Gemini, Codex) on detecting and repairing variability-induced compilation errors hidden under untested feature combinations in configurable C code.", "reason": "Task is restoring buildability of code under untested configurations, matching software_infrastructure's build-failure repair scope; single-activity study, not a field-wide survey.", "source": "crawl"}, {"paper": {"id": "2607.26656", "title": "Graph Is the Verifier: Agentic Reinforcement Learning for Interprocedural Vulnerability Detection", "authors": ["Yikun Li", "Ting Zhang", "Jiakun Liu", "Jinfeng Jiang", "Yuheng Yieh", "Yixin Yang", "Wen Bin Leow", "Yide Yin", "Yintong Huo", "Eng Lieh Ouh", "Lwin Khin Shar", "David Lo"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.26656", "github": "", "website": ""}}, "category": "software_security", "tags": [], "summary": "VulAgentRL is an agentic RL framework for interprocedural vulnerability detection that queries a Code Property Graph for cross-function evidence and uses exact-match graph verification as a training reward. It outperforms state-of-the-art detectors including frontier models under leakage-free evaluation.", "reason": "Agent gathers evidence and localizes vulnerabilities in code, matching software_security.", "source": "crawl"}, {"paper": {"id": "2607.27030", "title": "HoF-Bench: Rediscovering Real AI-Discovered CVEs Without Frontier Models", "authors": ["Petr Simecek", "Elnaz Babayeva", "Jiri Balhar", "Michal Bida", "Michal Buran", "Vaclav Cadek", "Luigino Camastra", "Tomas Dulka", "Michal Janocko", "Tomas Klohna", "Pavel Kohout", "Ondrej Kokes", "Adam Krivka", "Jakub Kubik", "Patrik Mada", "Igor Morgenstern", "Marek Pavelka", "Joshua Rogers", "Petr Stastny", "Jan Tattermusch", "Dmitrijs Trizna", "Martin Votruba", "Guido Vranken", "Jakub Zikl", "Evelina Gabasova", "Stanislav Fort"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27030", "github": "", "website": ""}}, "category": "software_security", "tags": ["benchmark"], "summary": "HoF-Bench is a benchmark of 95 real AI-discovered CVEs used to test whether minimal LLM-based vulnerability analyzers (without frontier models) can rediscover known vulnerabilities in mature open-source C/C++ projects. A detector-blinded judge grades findings on code path, root cause, condition, and impact.", "reason": "The served task is vulnerability detection/localization, placing it in software_security with a benchmark tag.", "source": "crawl"}, {"paper": {"id": "2601.05485", "title": "Readability-Robust Code Summarization via Meta Curriculum Learning", "authors": ["Wenhao Zeng", "Yitian Chai", "Hao Zhou", "Fandong Meng", "Jie Zhou", "Xiaodong Gu"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2601.05485", "github": "", "website": ""}}, "category": "software_comprehension", "tags": [], "summary": "Empirically shows LLMs like GPT-4o and DeepSeek-V3 degrade sharply on code summarization when code has poor readability, then proposes RoFTCodeSum, a curriculum-plus-meta-learning fine-tuning method to improve robustness against obfuscated code.", "reason": "Serves code comprehension/summarization task, which is the software_comprehension leaf.", "source": "crawl"}, {"paper": {"id": "2607.26451", "title": "ExplainBench: Evaluating Code Explanations from Agents", "authors": ["Zhiyuan Pan", "Sungmin Kang", "Imam Nur Bani Yusuf", "Abhik Roychoudhury"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.26451", "github": "", "website": ""}}, "category": "software_comprehension", "tags": ["benchmark"], "summary": "ExplainBench is a benchmark evaluating whether coding agents' natural-language explanations of patches accurately describe buggy code behavior and patch effects. It finds explanation quality diverges from SWE-bench rankings and proposes an audit agent to improve trustworthiness.", "reason": "Serves the task of understanding and documenting agent code changes, i.e. code comprehension/explanation, not judging change quality per se.", "source": "crawl"}, {"paper": {"id": "2607.26987", "title": "How Developers Experience Debugging Unfamiliar Codebases with Code Tours Generated and Evaluated by Local LLMs", "authors": ["Balfroid Martin", "Albert Julien", "Aliti Dzenatan", "Devroey Xavier", "Vanderose Benoît"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.26987", "github": "", "website": ""}}, "category": "software_comprehension", "tags": ["empirical"], "summary": "A user study of 26 developers debugging unfamiliar Java codebases using LLM-generated 'code tours' (onboarding documentation), examining trust calibration and preferences for tour style, and finding LLM-based tour evaluators are unreliable. The tours aid code comprehension to support debugging of unfamiliar repositories.", "reason": "An empirical study of a single tool for code documentation/comprehension, not a field-wide survey, so it routes to software_comprehension with the empirical tag rather than studies.", "source": "crawl"}]
```

</details>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Paper review 2026-07-30: 25 candidates #346

1. Kimi K3: Open Frontier Intelligence

2. Try Again, Don't Look Back: Blind Resampling Outperforms Self-Repair in Small Code Models

3. (Im)Paired Programming: Coding Agents Improve Productivity but Harm Understanding

4. Impossible to hide secret ...: Uncovering Security and Privacy Issues in LLM-native IDEs

5. Unpacking Vibe Coding: Help-Seeking Processes in Student-AI Interactions While Programming

6. A First Look at Coding Agents' Compliance with AI Contribution Rules in Open-Source Communities

7. Can AI agents conduct open-ended AI research? Early evidence from two case studies

8. REAP: Automatic Curation of Coding Agent Benchmarks from Interactive Production Usage

9. MultiFixer: A Coordinator-Proposer Based Multi-Agent Framework For Fixing Multi-Hunk Bugs

10. Cross-Model Cross-Language AI Coding Agent Performance: Accuracy and Speed of Parallel CLRS Algorithms

11. NISPO: Open-source IUPAC name generation tool

12. TraceCoder: Explainable and Auditable Code Generation with Position-Key Snippet Versioning

13. Better Call Grep: Evaluating and Improving Grep-Like Lexical Retrieval for Repository-Level Code Completion

14. DHRCL:Training Code LLMs with Dense Hierarchical Rewards and Curriculum Learning

15. Fewer Clarifications, Better Code: Benchmarking Cross-Session Personalized Ambiguity Adaptation in Coding Assistants

16. MRCoder: An Efficient Context Selecting Approach for Repository-Level Code Generation

17. VITAL-RAG: Invariance Race for Context Allocation in Coding Agents

18. MindForge: Teaching Small Language Models Whole-Life-Cycle Software Engineering via Source-Free Program Synthesis

19. SpecFirst: Behavioral Specification Elicitation as a First-Class Step in Agent-Based Program Synthesis from Scratch

20. An Empirical Study of Foundation Models for Variability-Induced Compilation Errors in Configurable C Code

21. Graph Is the Verifier: Agentic Reinforcement Learning for Interprocedural Vulnerability Detection

22. HoF-Bench: Rediscovering Real AI-Discovered CVEs Without Frontier Models

23. Readability-Robust Code Summarization via Meta Curriculum Learning

24. ExplainBench: Evaluating Code Explanations from Agents

25. How Developers Experience Debugging Unfamiliar Codebases with Code Tours Generated and Evaluated by Local LLMs

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Uh oh!

Paper review 2026-07-30: 25 candidates #346

Description

1. Kimi K3: Open Frontier Intelligence

2. Try Again, Don't Look Back: Blind Resampling Outperforms Self-Repair in Small Code Models

3. (Im)Paired Programming: Coding Agents Improve Productivity but Harm Understanding

4. Impossible to hide secret ...: Uncovering Security and Privacy Issues in LLM-native IDEs

5. Unpacking Vibe Coding: Help-Seeking Processes in Student-AI Interactions While Programming

6. A First Look at Coding Agents' Compliance with AI Contribution Rules in Open-Source Communities

7. Can AI agents conduct open-ended AI research? Early evidence from two case studies

8. REAP: Automatic Curation of Coding Agent Benchmarks from Interactive Production Usage

9. MultiFixer: A Coordinator-Proposer Based Multi-Agent Framework For Fixing Multi-Hunk Bugs

10. Cross-Model Cross-Language AI Coding Agent Performance: Accuracy and Speed of Parallel CLRS Algorithms

11. NISPO: Open-source IUPAC name generation tool

12. TraceCoder: Explainable and Auditable Code Generation with Position-Key Snippet Versioning

13. Better Call Grep: Evaluating and Improving Grep-Like Lexical Retrieval for Repository-Level Code Completion

14. DHRCL:Training Code LLMs with Dense Hierarchical Rewards and Curriculum Learning

15. Fewer Clarifications, Better Code: Benchmarking Cross-Session Personalized Ambiguity Adaptation in Coding Assistants

16. MRCoder: An Efficient Context Selecting Approach for Repository-Level Code Generation

17. VITAL-RAG: Invariance Race for Context Allocation in Coding Agents

18. MindForge: Teaching Small Language Models Whole-Life-Cycle Software Engineering via Source-Free Program Synthesis

19. SpecFirst: Behavioral Specification Elicitation as a First-Class Step in Agent-Based Program Synthesis from Scratch

20. An Empirical Study of Foundation Models for Variability-Induced Compilation Errors in Configurable C Code

21. Graph Is the Verifier: Agentic Reinforcement Learning for Interprocedural Vulnerability Detection

22. HoF-Bench: Rediscovering Real AI-Discovered CVEs Without Frontier Models

23. Readability-Robust Code Summarization via Meta Curriculum Learning

24. ExplainBench: Evaluating Code Explanations from Agents

25. How Developers Experience Debugging Unfamiliar Codebases with Code Tours Generated and Evaluated by Local LLMs

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions