[{"paper": {"id": "2605.26494", "title": "The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence", "authors": ["Aili Chen", "Aonian Li", "Baichuan Zhou", "Bangwei Gong", "Binyang Jiang", "Boji Dan", "Changhao Zhang", "Changqing Yu", "Chao Wang", "Cheng Ma", "Cheng Zhong", "Cheng Zhu", "Chengjun Xiao", "Chengyi Yang", "Chengyu Du", "Chenyang Zhang", "Chi Zhang", "Chuangyi Huang", "Chunhao Zhang", "Chunhui Du", "Chunyu Zhao", "Congchao Guo", "Da Chen", "Deming Ding", "Dianjun Sun", "Dong Li", "Dongyu Zhang", "Enhui Yang", "Fei Yu", "Guang Zheng", "Guodong Zheng", "Guohong Li", "Haichao Zhu", "Haigang Zhou", "Haimo Zhang", "Han Ding", "Hao Zhang", "Haohai Sun", "Haolin Lyu", "Haonan Lu", "Haoyu Wang", "Huajie Shi", "Huiyang Li", "Jiacheng Chen", "Jian Zhang", "Jiaqi Zhuang", "Jiaren Cai", "Jiaxin Pan", "Jiayao Li", "Jiayuan Song", "Jichuan Zhang", "Jie Wang", "Jihao Gu", "Jin Zhu", "Jingwei Dong", "Jingyang Li", "Jingyu Zhang", "Jingze Zhuang", "Jinhao Tian", "Jinli Liu", "Jinyi Hu", "Jun Tao", "Jun Zhang", "Junbin Ruan", "Junhao Xu", "Junjie Yan", "Junteng Liu", "Junxian He", "Kang Xu", "Ke Ji", "Ke Yang", "Kecheng Xiao", "Keyu Duan", "Keyu Li", "Le Han", "Letian Ruan", "Li Yuan", "Lianfei Yu", "Liheng Feng", "Lijie Mo", "Lin Li", "Linge Du", "Lingye Bao", "Lingyu Yang", "Lingyuan Zhou", "Loki", "Lu Chen", "Lunbin Zeng", "Ming Li", "Ming Zhong", "Mingliang Tao", "Mingyuan Chi", "Mujie Lin", "Nan Hu", "Ningxin Chen", "Peiyin Zhu", "Peng Gao", "Pengcheng Gao", "Pengfei Li", "Penglin Li", "Pengyu Zhao", "Qibin Ren", "Qibing Ren", "Qidi Xu", "Qihan Ren", "Qile Li", "Qin Wang", "Quanliang Chen", "Qunhong Zeng", "Rong Tian", "Rongxin Guo", "Rui Dong", "Ruitao Leng", "Ruize Zhang", "Shanqi Liu", "Shaoxiang Chen", "Shaoyu Chen", "Sheng Jia", "Shun Yao", "Shuoran Zhao", "Shuqi Yu", "Sichen Li", "Sicheng Pan", "Songquan Zhu", "Tengfei Li", "Tian Xie", "Tiancheng Qin", "Tianle Li", "Tianrun Liang", "Wei Liu", "Weiqi Xu", "Weitao Li", "Weixiang Chen", "Weiyu Cheng", "Weiyu Zhang", "Wenhu Chen", "Wenqian Zhao", "Xiancai Chen", "Xiangjun Song", "Xiangyuan Wang", "Xianzhen Luo", "Xiao Luo", "Xiao Su", "Xiaobo Li", "Xiaodong Han", "Xiaojie Wu", "Xihao Song", "Xingyi Han", "Xinyu Guan", "Xuan Lu", "Xun Zou", "Xunhao Lai", "Xutong Li", "Xuyang Shen", "Yan Gong", "Yan Ma", "Yang Jiao", "Yang Wang", "Yang Xu", "Yangsen Wang", "Ye Tang", "Yicheng Chen", "Yihang Wang", "Yinran Qiu", "Yiqi Shi", "Yiting Guo", "Yiwen Huang", "Yixuan Wang", "Yongyi Hu", "Yu Gao", "Yu Zhang", "Yuan Li", "Yuanxiang Ying", "Yuanzhen Zhang", "Yubo Wang", "Yuchen Song", "Yufeng Yang", "Yuhang Meng", "Yuhang Miao", "Yuhao Li", "Yujie Liu", "Yulin Hu", "Yunan Huang", "Yunji Li", "Yunyi Huang", "Yusen Zhang", "Yusu Hong", "Yutao Xie", "Yutong Zhang", "Yuwen Liao", "Yuxuan Shi", "Yuze Wenren", "Zebin Li", "Zehan Li", "Zejian Luo", "Zeyu Jin", "Zeyuan Sun", "Zhanpeng Zhou", "Zhaochen Su", "Zhendong Li", "Zhengmao Zhu", "Zhengyuan Peng", "Zhenhua Fan", "Zhi Zhang", "Zhichao Xu", "Zhiheng Lv", "Zhikang Xu", "Zhitao He", "Zhiwei He", "Zhongyuan Li", "Zibo Gao", "Zijia Wu", "Zijian Song", "Zijian Zhou", "Zijun Sun", "Zishan Huang", "Ziying Chen", "Ziyue Ge"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2605.26494", "github": "", "website": ""}}, "category": "foundation_models", "tags": ["model"], "summary": "Introduces the MiniMax-M2 series, a sparse MoE foundation model (229.9B total/9.8B active params) built end-to-end for agentic deployment across agentic coding, deep search, office tasks, and reasoning via agent-driven data pipelines and a scalable RL system.", "reason": "A flagship general-purpose agentic model spanning multiple activities and worlds, matching the foundation_models definition.", "source": "crawl"}, {"paper": {"id": "2603.26487", "title": "To Ban or not to Ban? How Open Source Projects Govern GenAI Contributions", "authors": ["Wenhao Yang", "Runzhi He", "Minghui Zhou"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2603.26487", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "An empirical, multi-stage qualitative analysis of 67 OSS projects examining how they govern GenAI contributions to code, issues, reviews, and security reports, identifying governance orientations and strategies.", "reason": "Field-wide empirical study of community practices around AI-generated contributions, matching the studies leaf's adoption/behavioral-study scope rather than proposing a task-performing agent.", "source": "crawl"}, {"paper": {"id": "2607.27250", "title": "Do Context Files Help Coding Agents? A Two-Agent Ablation Study on Real Repositories", "authors": ["Prakhar Khatri"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-28", "links": {"paper": "https://arxiv.org/abs/2607.27250", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "A controlled ablation study across Claude Code and Codex on 17 real repository tasks finds persistent context files (AGENTS.md/CLAUDE.md) do not measurably improve correctness, with failures traced to implementation-skill gaps rather than missing repo knowledge.", "reason": "An empirical/behavioral study of how coding agents use context, not a proposed method or agent — object is the agents themselves.", "source": "crawl"}, {"paper": {"id": "2607.27923", "title": "The Case for Vibe Modeling: A Missing Step in AI-Based Trustworthy Software Development", "authors": ["Shalini Chakraborty", "Michael Mittermaier", "Judith Michael"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.27923", "github": "", "website": ""}}, "category": "studies", "tags": ["position"], "summary": "A position paper arguing that AI-based software development should include 'vibe modeling', an intermediate abstraction layer between natural-language prompts and code generation, supported by a student survey on trust and validation. It proposes no task-performing agent, focusing instead on the practice of AI-assisted development.", "reason": "Its object is the practice/process of AI-based software development itself, matching the studies/position boundary.", "source": "crawl"}, {"paper": {"id": "2607.27249", "title": "OwlPath: Lossless Knowledge Compression for LLM Bug Repair", "authors": ["Bo Zhang", "Ren Pan", "Huan Chen", "Xiang Song"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-28", "links": {"paper": "https://arxiv.org/abs/2607.27249", "github": "", "website": ""}}, "category": "software_debugging", "tags": [], "summary": "OwlPath is an OWL2 ontology-based reasoning layer atop CodeGraph that enables structural, multi-hop code retrieval to help LLM bug-repair agents resolve SWE-bench Pro issues with fewer tokens and higher recall than plain-text retrieval baselines.", "reason": "Serves issue-resolution agents by improving structural code retrieval for bug repair (software_debugging).", "source": "crawl"}, {"paper": {"id": "2607.28587", "title": "PAIChecker: Uncovering and Checking PR-Issue Misalignment in SWE-Bench-Like Benchmarks", "authors": ["Manyi Wang", "Junjielong Xu", "Pinjia He"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28587", "github": "", "website": ""}}, "category": "software_debugging", "tags": [], "summary": "PAIChecker is a multi-agent system that detects PR-Issue misalignment in SWE-bench-like benchmarks, finding 13.6% of SWE-bench Verified instances misaligned and achieving up to 92% detection accuracy across benchmarks.", "reason": "Benchmark routing: it validates and improves issue-resolution (SWE-bench style) benchmark construction, serving software_debugging.", "source": "crawl"}, {"paper": {"id": "2607.27271", "title": "RLPF: Reinforcement Learning from Performance Feedback for Code Generation", "authors": ["Huihao Jing", "Haozhe Cui", "Wenbin Hu", "Shaojin Chen", "Haochen Shi", "Changxuan Fan", "Yuxuan Liu", "Hanyu Yang", "Sirui Zhang", "Ziyi Chen", "Haoran Li", "Yangqiu Song"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27271", "github": "", "website": ""}}, "category": "software_code_generation", "tags": ["model"], "summary": "RLPF trains code-generation models with staged reinforcement learning rewards from execution progress and runtime improvement, teaching agents to produce not just correct but also efficient code, substantially improving Qwen3-32B on PerfCodeBench.", "reason": "Serves the code-generation task by optimizing for correctness and efficiency of newly produced programs.", "source": "crawl"}, {"paper": {"id": "2607.27877", "title": "An Empirical Study of Coordination Mode as the First-Class Citizen in From-Scratch Multi-Agent Coding", "authors": ["Yanyu Ren", "Yunfeng Bai", "Xizheng Wang", "Li Chen", "Dan Li"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.27877", "github": "", "website": ""}}, "category": "software_code_generation", "tags": ["benchmark", "empirical"], "summary": "MSEval is a benchmark and execution engine (LegoGent) for evaluating multi-agent teams building real-world full-stack software from scratch across 10 collaboration topologies, measuring functional success, latency, and cost. It finds coordination topology rivals model capability in shaping quality-cost-speed tradeoffs.", "reason": "Proposes a benchmark for multi-agent from-scratch application generation, so it routes to the code-generation task leaf per the benchmark-dominance rule rather than studies.", "source": "crawl"}, {"paper": {"id": "2607.28027", "title": "VISA: A Structured Description Protocol for Agent-Based Simulation Models Towards Machine Reproducibility", "authors": ["Zhou He"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28027", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "VISA is a structured, symbol-based description protocol for agent-based simulation models that enables machine-parseable specifications, consistency checking, and LLM-driven code generation to reproduce models across platforms (e.g., NetLogo to Python). It validates the approach by reproducing three independently authored models via its author-check-code-reproduce loop.", "reason": "The protocol's ultimate purpose is producing executable code that reproduces a specified simulation model, matching the artifact/code-generation leaf.", "source": "crawl"}, {"paper": {"id": "2607.28086", "title": "Distilling Answer Set Programming Theories from Large Language Models", "authors": ["Nelson Higuera Ruiz", "Markus Hofmarcher", "Claudiu Leoveanu-Condrei"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28086", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "This work studies whether LLMs, guided by a fixed agent harness with an ASP solver in the loop, can distill complete and correct Answer Set Programming theories from scratch, evaluated on VQA benchmarks (CLEVR, GQA, CLEVRER) across nine models. Frontier models reach near-100% accuracy on several benchmarks.", "reason": "An agent harness produces logic-program code (ASP theories) as the deliverable, matching code generation in the artifact branch.", "source": "crawl"}, {"paper": {"id": "2604.11270", "title": "Evaluating LLM Agents on Automated Software Analysis Tasks", "authors": ["Islem Bouzenia", "Cristian Cadar", "Michael Pradel"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2604.11270", "github": "", "website": ""}}, "category": "software_infrastructure", "tags": ["benchmark"], "summary": "Introduces AnalysisBench and AnalysisAgent to evaluate LLM agents on installing, configuring, and running third-party software analysis tools alongside target C/C++/Java projects, outperforming baseline agent architectures. Extended runs also uncover previously unknown defects in real tools.", "reason": "The task is enabling work—setting up and configuring tools/environments so analysis can run—matching software_infrastructure.", "source": "crawl"}, {"paper": {"id": "2607.27674", "title": "Can Large Language Models Resolve Real Java Merge Conflicts? An Evaluation with a Calibrated LLM-as-Judge", "authors": ["Bowen Shen"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.27674", "github": "", "website": ""}}, "category": "software_infrastructure", "tags": [], "summary": "This paper evaluates whether LLM agents can resolve real Java merge conflicts, building a generate-validate-retry solver and a calibrated LLM-as-judge metric validated against human labels. LLM resolutions match developer intent on ~55% of true conflicts, outperforming traditional tools mainly through higher coverage.", "reason": "Merge-conflict resolution is explicitly a version-control operation under software_infrastructure.", "source": "crawl"}, {"paper": {"id": "2607.28591", "title": "Change2Task: From Repository Changes to Executable Coding Agent Tasks and Environments", "authors": ["Haomin Qi", "Xingliang Wang", "Xuanqi Gao", "Baihui Sang", "Xin Zhang", "Minghua Ma", "Pengfei Gao", "Yu Kang", "Qingwei Lin", "Saravan Rajmohan", "Dongmei Zhang", "Qi Zhang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28591", "github": "", "website": ""}}, "category": "software_infrastructure", "tags": ["training-data"], "summary": "Change2Task converts merged pull requests from repository history into verified, executable coding-agent tasks and environments spanning bug fixes, feature addition, test generation, API migration, and security repair, reducing environment setup effort for agent training and evaluation.", "reason": "A resource paper whose contribution is constructing executable tasks and environments (enabling work) to train/evaluate coding agents across multiple lifecycle activities, matching software_infrastructure's enabling-work scope.", "source": "crawl"}, {"paper": {"id": "2607.27409", "title": "SWE-NFI: Studying and Benchmarking Coding Agents for Non-Functional Improvements", "authors": ["Pengyu Xue", "He Yang Yuan", "Xin Wang", "Junkai Chen", "Haonan Zhang", "Boyuan Chen", "Zishuo Ding", "Zhenhao Li", "Weiyi Shang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27409", "github": "", "website": ""}}, "category": "software_maintenance", "tags": ["benchmark"], "summary": "SWE-NFI benchmarks coding agents on behavior-preserving non-functional improvements (e.g., structural code quality) using 188 tasks from real merged PRs, evaluated with 92 executable rules. State-of-the-art agents lag human developers substantially on these non-functional improvement tasks.", "reason": "Evaluates agents on behavior-preserving code improvements, matching software_maintenance's refactoring/evolution scope.", "source": "crawl"}, {"paper": {"id": "2503.15669", "title": "ECO: An LLM-Driven Efficient Code Optimizer for Warehouse Scale Computers", "authors": ["Hannah Lin", "Martin Maas", "Maximilian Roquemore", "Arman Hasanzadeh", "Fred Lewis", "Yusuf Simonson", "Tzu-Wei Yang", "Amir Yazdanbakhsh", "Deniz Altinbüken", "Florin Papa", "Maggie Nolan Edmonds", "Aditya Patil", "Don Schwarz", "Satish Chandra", "Chris Kennelly", "Milad Hashemi", "Parthasarathy Ranganathan"], "venue": "arXiv 2025/03", "category": "", "published": "2025-03-19", "links": {"paper": "https://arxiv.org/abs/2503.15669", "github": "", "website": ""}}, "category": "software_maintenance", "tags": [], "summary": "ECO is an LLM-driven system deployed at Google that mines historical commits for performance anti-patterns, then automatically refactors matching code segments across a billion-line codebase, verifies, reviews, and measures production impact.", "reason": "Behavior-preserving performance optimization of existing code across a large codebase fits software_maintenance.", "source": "crawl"}, {"paper": {"id": "2607.27228", "title": "AI-assisted pre-review of open-source software submissions: an experience report from BOSC 2026", "authors": ["Tazro Ohta", "Nomi L. Harris", "Seth Carbon"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-14", "links": {"paper": "https://arxiv.org/abs/2607.27228", "github": "", "website": ""}}, "category": "software_review", "tags": [], "summary": "Describes bosc-pre-review, an agentic skill paired with Runabilly (which builds/tests submitted projects in Docker) to pre-assess open-source software submissions against a conference review rubric, with humans making final decisions.", "reason": "The agent judges submitted code/projects for conference review, matching software_review's patch/change-quality judgment scope.", "source": "crawl"}, {"paper": {"id": "2607.27528", "title": "ThreatForest: Multi-Agent Attack Tree Generation with Pluggable TTP Framework Mapping", "authors": ["Cristian Leo", "Anton Dykyi", "Danny Cortegaca", "Daniel Begimher", "Prakash Jha"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27528", "github": "", "website": ""}}, "category": "software_security", "tags": [], "summary": "Presents ThreatForest, a multi-agent pipeline that analyzes source code repositories to generate structured attack trees, maps steps to MITRE ATT&CK/CAPEC TTPs, and synthesizes mitigations.", "reason": "Repository auditing for threat modeling and mitigation generation is security-assurance work on code, matching software_security's repository-auditing scope.", "source": "crawl"}, {"paper": {"id": "2607.27359", "title": "PROGRESS: Property-Guided Regression Search for Semantic Falsification", "authors": ["Davis Tocheuk Mo", "Noshin Ulfat", "Matthew B. Dwyer", "Soneya Binta Hossain"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27359", "github": "", "website": ""}}, "category": "software_testing", "tags": [], "summary": "PROGRESS combines LLM-generated intent-driven properties with coverage-guided evolutionary regression-test search to expose semantic bugs missed by regression-derived oracles. Evaluated on 25 Java systems, it detects far more real bugs than regression-test generation or standalone property-based testing.", "reason": "Proactive test/property generation for correctness assurance independent of a reported defect fits software_testing.", "source": "crawl"}, {"paper": {"id": "2607.28271", "title": "Agentic Method for Deterministic Validation of Legacy Code Migration", "authors": ["Andras Ferenczi", "Jordan Docherty", "Mariya Bessonov", "Matthew Findlay", "Krishna Lingamneni"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28271", "github": "", "website": ""}}, "category": "software_testing", "tags": [], "summary": "Proposes 'Locksmith Loop,' an agentic test-synthesis method that runs instrumented COBOL and migrated Java side-by-side, iteratively searching inputs and mutations to validate migration correctness via coverage and deterministic parity checks. Evaluated on three COBOL-to-Java case studies achieving near-complete branch coverage.", "reason": "The task is proactive test/witness generation to validate correctness of migrated code, matching software_testing's coverage and verification focus.", "source": "crawl"}, {"paper": {"id": "2607.27235", "title": "RadHarmony: Radiological Data Handling in the Era of Agentic AI", "authors": ["Frank Li", "Bardia Khosravi", "Mohammadreza Chavoshi", "Theo Dapamede", "YoungSeok Jeon", "Janice Newsome", "Hari Trivedi", "Judy Gichoya"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.27235", "github": "", "website": ""}}, "category": "software_development", "tags": [], "summary": "RadHarmony is a Python library unifying heterogeneous radiological datasets, including an AI-agent skill that guides integrating new datasets into the library end-to-end, from raw data inspection through code generation and testing.", "reason": "The agent implements new dataset-integration functionality inside an existing codebase, matching feature development into an existing repository.", "source": "crawl"}, {"paper": {"id": "2607.27380", "title": "VideoCoCo: Code-as-CoT for Physically-Consistent Video Generation via an Agentic Dual-Engine System", "authors": ["Haodong Li", "Tianfei Ren", "Xiaoxiao Ma", "Chunmei Qing", "Zhen Fang", "Sipeng He", "Ziyu Guo", "Haoyu Wu", "Juanxi Tian", "Yihang Zou", "Ruichuan An", "Dongzhi Jiang", "Boxue Yang", "Ji Xie", "Xu Huang", "Wenhao Yan", "Jialv Zou", "Zhengrong Yue", "Yaxin Luo", "Xiaotong Li", "Yuzhu Wang", "Junyan Ye", "Jinjing Zhao", "Zehui Chen", "Lin Chen", "Renye Yan", "Feng Zhao", "Pheng-Ann Heng"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27380", "github": "", "website": ""}}, "category": "cad", "tags": ["training-data"], "summary": "VideoCoCo uses a coding agent to synthesize executable Blender programs that construct a scene's spatiotemporal evolution as a process-level chain of thought, which is rendered into a draft and then refined into photorealistic video. A curated dataset (VideoCoCo-3K) is released to adapt the video editor to these simulated drafts.", "reason": "A coding agent produces executable 3D-scene code (Blender program) as the mechanism, matching CAD's 3D-scene-construction-as-code inclusion.", "source": "crawl"}, {"paper": {"id": "2607.27558", "title": "Drawing-Recode: Annotation Grounding for Parametric CAD Code Generation from Raster 2D CAD Drawings", "authors": ["Mingi Kim", "Yongjun Kim", "Hyungki Kim"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.27558", "github": "", "website": ""}}, "category": "cad", "tags": [], "summary": "Drawing-Recode recovers parametric CAD sequences as structured CAD code from raster 2D CAD drawings, grounding dimensional annotations to geometry via cross-attention and a new Annotation Grounding Loss. An LLM generates the final CAD code, improving over baselines even on scanned industrial drawings.", "reason": "Delivers CAD programs as code from drawings, matching the cad leaf's definition of producing 3D/CAD code artifacts.", "source": "crawl"}, {"paper": {"id": "2607.28050", "title": "IndustryForge-27B: A Domain-Enhanced Multimodal Foundation Model for Industrial CAD", "authors": ["Nianchen Deng", "Jiaxin Ai", "Tao Hu", "Shu Zou", "Yurui Dong", "Siqi Li", "Xinyu Cai", "Xuemeng Yang", "Licheng Wen", "Hongbin Zhou", "Hairong Zhang", "Pinlong Cai", "Botian Shi"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28050", "github": "", "website": ""}}, "category": "cad", "tags": ["model"], "summary": "IndustryForge-27B is a multimodal foundation model fine-tuned on industrial CAD corpora to generate parametric CAD code and COM API scripts from engineering drawings and geometry. It outperforms GPT-5.4 on CAD benchmarks while retaining general capability.", "reason": "Specialized model for producing CAD/parametric code, not a general-purpose flagship model, so routes to the cad artifact leaf plus model tag.", "source": "crawl"}, {"paper": {"id": "2607.28033", "title": "DataClawEval: A Benchmark for Data Engineering Agents in Real Industrial Harness", "authors": ["Debin Meng", "Jiaming Yang", "Zefang Zong", "Tengyue Xu", "Haining Xie", "Yang Li", "Peng Chen"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28033", "github": "", "website": ""}}, "category": "database", "tags": ["benchmark"], "summary": "DataClawEval is a benchmark of 100 production-grade data engineering tasks spanning PySpark, MySQL, HiveSQL, Presto/Trino and FlinkSQL, graded deterministically in sandboxed environments. It evaluates 16 frontier agents, showing best performance of only 74.9 and strong domain specialization across engines.", "reason": "Tasks center on writing/executing SQL and data-pipeline code across engines, closest to the database artifact leaf.", "source": "crawl"}, {"paper": {"id": "2607.27699", "title": "RefineSVG: Visual Feedback-Driven Reinforcement Learning for Image-to-SVG Generation", "authors": ["Shaobo Liu", "Feiqiao Mao", "Shuaishuai Zhou", "Yan Zhan", "Weiqi Tan", "Zhiqiong Lu", "Zhengping Liang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.27699", "github": "", "website": ""}}, "category": "graphics", "tags": [], "summary": "RefineSVG is a closed-loop visual-feedback framework where an MLLM generates SVG code, renders it, compares against the target image via a visual residual map, and iteratively self-corrects using RL-trained agentic refinement. It improves reconstruction fidelity and code efficiency over open-loop SVG generation.", "reason": "The deliverable is vector-graphic (SVG) code produced from an image spec, matching artifact/graphics.", "source": "crawl"}]
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 135 out-of-scope; 0 failed (retried next run).
1. The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence
Aili Chen, Aonian Li, Baichuan Zhou, et al. · arXiv 2026/07 · paper
proposed:
foundation_models· tags:model2. To Ban or not to Ban? How Open Source Projects Govern GenAI Contributions
Wenhao Yang, Runzhi He, Minghui Zhou · arXiv 2026/07 · paper
proposed:
studies· tags:empirical3. Do Context Files Help Coding Agents? A Two-Agent Ablation Study on Real Repositories
Prakhar Khatri · arXiv 2026/07 · paper
proposed:
studies· tags:empirical4. The Case for Vibe Modeling: A Missing Step in AI-Based Trustworthy Software Development
Shalini Chakraborty, Michael Mittermaier, Judith Michael · arXiv 2026/07 · paper
proposed:
studies· tags:position5. OwlPath: Lossless Knowledge Compression for LLM Bug Repair
Bo Zhang, Ren Pan, Huan Chen, et al. · arXiv 2026/07 · paper
proposed:
software_debugging· tags: none6. PAIChecker: Uncovering and Checking PR-Issue Misalignment in SWE-Bench-Like Benchmarks
Manyi Wang, Junjielong Xu, Pinjia He · arXiv 2026/07 · paper
proposed:
software_debugging· tags: none7. RLPF: Reinforcement Learning from Performance Feedback for Code Generation
Huihao Jing, Haozhe Cui, Wenbin Hu, et al. · arXiv 2026/07 · paper
proposed:
software_code_generation· tags:model8. An Empirical Study of Coordination Mode as the First-Class Citizen in From-Scratch Multi-Agent Coding
Yanyu Ren, Yunfeng Bai, Xizheng Wang, et al. · arXiv 2026/07 · paper
proposed:
software_code_generation· tags:benchmarkempirical9. VISA: A Structured Description Protocol for Agent-Based Simulation Models Towards Machine Reproducibility
Zhou He · arXiv 2026/07 · paper
proposed:
software_code_generation· tags: none10. Distilling Answer Set Programming Theories from Large Language Models
Nelson Higuera Ruiz, Markus Hofmarcher, Claudiu Leoveanu-Condrei · arXiv 2026/07 · paper
proposed:
software_code_generation· tags: none11. Evaluating LLM Agents on Automated Software Analysis Tasks
Islem Bouzenia, Cristian Cadar, Michael Pradel · arXiv 2026/07 · paper
proposed:
software_infrastructure· tags:benchmark12. Can Large Language Models Resolve Real Java Merge Conflicts? An Evaluation with a Calibrated LLM-as-Judge
Bowen Shen · arXiv 2026/07 · paper
proposed:
software_infrastructure· tags: none13. Change2Task: From Repository Changes to Executable Coding Agent Tasks and Environments
Haomin Qi, Xingliang Wang, Xuanqi Gao, et al. · arXiv 2026/07 · paper
proposed:
software_infrastructure· tags:training-data14. SWE-NFI: Studying and Benchmarking Coding Agents for Non-Functional Improvements
Pengyu Xue, He Yang Yuan, Xin Wang, et al. · arXiv 2026/07 · paper
proposed:
software_maintenance· tags:benchmark15. ECO: An LLM-Driven Efficient Code Optimizer for Warehouse Scale Computers
Hannah Lin, Martin Maas, Maximilian Roquemore, et al. · arXiv 2025/03 · paper
proposed:
software_maintenance· tags: none16. AI-assisted pre-review of open-source software submissions: an experience report from BOSC 2026
Tazro Ohta, Nomi L. Harris, Seth Carbon · arXiv 2026/07 · paper
proposed:
software_review· tags: none17. ThreatForest: Multi-Agent Attack Tree Generation with Pluggable TTP Framework Mapping
Cristian Leo, Anton Dykyi, Danny Cortegaca, et al. · arXiv 2026/07 · paper
proposed:
software_security· tags: none18. PROGRESS: Property-Guided Regression Search for Semantic Falsification
Davis Tocheuk Mo, Noshin Ulfat, Matthew B. Dwyer, et al. · arXiv 2026/07 · paper
proposed:
software_testing· tags: none19. Agentic Method for Deterministic Validation of Legacy Code Migration
Andras Ferenczi, Jordan Docherty, Mariya Bessonov, et al. · arXiv 2026/07 · paper
proposed:
software_testing· tags: none20. RadHarmony: Radiological Data Handling in the Era of Agentic AI
Frank Li, Bardia Khosravi, Mohammadreza Chavoshi, et al. · arXiv 2026/07 · paper
proposed:
software_development· tags: none21. VideoCoCo: Code-as-CoT for Physically-Consistent Video Generation via an Agentic Dual-Engine System
Haodong Li, Tianfei Ren, Xiaoxiao Ma, et al. · arXiv 2026/07 · paper
proposed:
cad· tags:training-data22. Drawing-Recode: Annotation Grounding for Parametric CAD Code Generation from Raster 2D CAD Drawings
Mingi Kim, Yongjun Kim, Hyungki Kim · arXiv 2026/07 · paper
proposed:
cad· tags: none23. IndustryForge-27B: A Domain-Enhanced Multimodal Foundation Model for Industrial CAD
Nianchen Deng, Jiaxin Ai, Tao Hu, et al. · arXiv 2026/07 · paper
proposed:
cad· tags:model24. DataClawEval: A Benchmark for Data Engineering Agents in Real Industrial Harness
Debin Meng, Jiaming Yang, Zefang Zong, et al. · arXiv 2026/07 · paper
proposed:
database· tags:benchmark25. RefineSVG: Visual Feedback-Driven Reinforcement Learning for Image-to-SVG Generation
Shaobo Liu, Feiqiao Mao, Shuaishuai Zhou, et al. · arXiv 2026/07 · paper
proposed:
graphics· tags: nonemachine payload (do not edit)
[{"paper": {"id": "2605.26494", "title": "The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence", "authors": ["Aili Chen", "Aonian Li", "Baichuan Zhou", "Bangwei Gong", "Binyang Jiang", "Boji Dan", "Changhao Zhang", "Changqing Yu", "Chao Wang", "Cheng Ma", "Cheng Zhong", "Cheng Zhu", "Chengjun Xiao", "Chengyi Yang", "Chengyu Du", "Chenyang Zhang", "Chi Zhang", "Chuangyi Huang", "Chunhao Zhang", "Chunhui Du", "Chunyu Zhao", "Congchao Guo", "Da Chen", "Deming Ding", "Dianjun Sun", "Dong Li", "Dongyu Zhang", "Enhui Yang", "Fei Yu", "Guang Zheng", "Guodong Zheng", "Guohong Li", "Haichao Zhu", "Haigang Zhou", "Haimo Zhang", "Han Ding", "Hao Zhang", "Haohai Sun", "Haolin Lyu", "Haonan Lu", "Haoyu Wang", "Huajie Shi", "Huiyang Li", "Jiacheng Chen", "Jian Zhang", "Jiaqi Zhuang", "Jiaren Cai", "Jiaxin Pan", "Jiayao Li", "Jiayuan Song", "Jichuan Zhang", "Jie Wang", "Jihao Gu", "Jin Zhu", "Jingwei Dong", "Jingyang Li", "Jingyu Zhang", "Jingze Zhuang", "Jinhao Tian", "Jinli Liu", "Jinyi Hu", "Jun Tao", "Jun Zhang", "Junbin Ruan", "Junhao Xu", "Junjie Yan", "Junteng Liu", "Junxian He", "Kang Xu", "Ke Ji", "Ke Yang", "Kecheng Xiao", "Keyu Duan", "Keyu Li", "Le Han", "Letian Ruan", "Li Yuan", "Lianfei Yu", "Liheng Feng", "Lijie Mo", "Lin Li", "Linge Du", "Lingye Bao", "Lingyu Yang", "Lingyuan Zhou", "Loki", "Lu Chen", "Lunbin Zeng", "Ming Li", "Ming Zhong", "Mingliang Tao", "Mingyuan Chi", "Mujie Lin", "Nan Hu", "Ningxin Chen", "Peiyin Zhu", "Peng Gao", "Pengcheng Gao", "Pengfei Li", "Penglin Li", "Pengyu Zhao", "Qibin Ren", "Qibing Ren", "Qidi Xu", "Qihan Ren", "Qile Li", "Qin Wang", "Quanliang Chen", "Qunhong Zeng", "Rong Tian", "Rongxin Guo", "Rui Dong", "Ruitao Leng", "Ruize Zhang", "Shanqi Liu", "Shaoxiang Chen", "Shaoyu Chen", "Sheng Jia", "Shun Yao", "Shuoran Zhao", "Shuqi Yu", "Sichen Li", "Sicheng Pan", "Songquan Zhu", "Tengfei Li", "Tian Xie", "Tiancheng Qin", "Tianle Li", "Tianrun Liang", "Wei Liu", "Weiqi Xu", "Weitao Li", "Weixiang Chen", "Weiyu Cheng", "Weiyu Zhang", "Wenhu Chen", "Wenqian Zhao", "Xiancai Chen", "Xiangjun Song", "Xiangyuan Wang", "Xianzhen Luo", "Xiao Luo", "Xiao Su", "Xiaobo Li", "Xiaodong Han", "Xiaojie Wu", "Xihao Song", "Xingyi Han", "Xinyu Guan", "Xuan Lu", "Xun Zou", "Xunhao Lai", "Xutong Li", "Xuyang Shen", "Yan Gong", "Yan Ma", "Yang Jiao", "Yang Wang", "Yang Xu", "Yangsen Wang", "Ye Tang", "Yicheng Chen", "Yihang Wang", "Yinran Qiu", "Yiqi Shi", "Yiting Guo", "Yiwen Huang", "Yixuan Wang", "Yongyi Hu", "Yu Gao", "Yu Zhang", "Yuan Li", "Yuanxiang Ying", "Yuanzhen Zhang", "Yubo Wang", "Yuchen Song", "Yufeng Yang", "Yuhang Meng", "Yuhang Miao", "Yuhao Li", "Yujie Liu", "Yulin Hu", "Yunan Huang", "Yunji Li", "Yunyi Huang", "Yusen Zhang", "Yusu Hong", "Yutao Xie", "Yutong Zhang", "Yuwen Liao", "Yuxuan Shi", "Yuze Wenren", "Zebin Li", "Zehan Li", "Zejian Luo", "Zeyu Jin", "Zeyuan Sun", "Zhanpeng Zhou", "Zhaochen Su", "Zhendong Li", "Zhengmao Zhu", "Zhengyuan Peng", "Zhenhua Fan", "Zhi Zhang", "Zhichao Xu", "Zhiheng Lv", "Zhikang Xu", "Zhitao He", "Zhiwei He", "Zhongyuan Li", "Zibo Gao", "Zijia Wu", "Zijian Song", "Zijian Zhou", "Zijun Sun", "Zishan Huang", "Ziying Chen", "Ziyue Ge"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2605.26494", "github": "", "website": ""}}, "category": "foundation_models", "tags": ["model"], "summary": "Introduces the MiniMax-M2 series, a sparse MoE foundation model (229.9B total/9.8B active params) built end-to-end for agentic deployment across agentic coding, deep search, office tasks, and reasoning via agent-driven data pipelines and a scalable RL system.", "reason": "A flagship general-purpose agentic model spanning multiple activities and worlds, matching the foundation_models definition.", "source": "crawl"}, {"paper": {"id": "2603.26487", "title": "To Ban or not to Ban? How Open Source Projects Govern GenAI Contributions", "authors": ["Wenhao Yang", "Runzhi He", "Minghui Zhou"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2603.26487", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "An empirical, multi-stage qualitative analysis of 67 OSS projects examining how they govern GenAI contributions to code, issues, reviews, and security reports, identifying governance orientations and strategies.", "reason": "Field-wide empirical study of community practices around AI-generated contributions, matching the studies leaf's adoption/behavioral-study scope rather than proposing a task-performing agent.", "source": "crawl"}, {"paper": {"id": "2607.27250", "title": "Do Context Files Help Coding Agents? A Two-Agent Ablation Study on Real Repositories", "authors": ["Prakhar Khatri"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-28", "links": {"paper": "https://arxiv.org/abs/2607.27250", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "A controlled ablation study across Claude Code and Codex on 17 real repository tasks finds persistent context files (AGENTS.md/CLAUDE.md) do not measurably improve correctness, with failures traced to implementation-skill gaps rather than missing repo knowledge.", "reason": "An empirical/behavioral study of how coding agents use context, not a proposed method or agent — object is the agents themselves.", "source": "crawl"}, {"paper": {"id": "2607.27923", "title": "The Case for Vibe Modeling: A Missing Step in AI-Based Trustworthy Software Development", "authors": ["Shalini Chakraborty", "Michael Mittermaier", "Judith Michael"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.27923", "github": "", "website": ""}}, "category": "studies", "tags": ["position"], "summary": "A position paper arguing that AI-based software development should include 'vibe modeling', an intermediate abstraction layer between natural-language prompts and code generation, supported by a student survey on trust and validation. It proposes no task-performing agent, focusing instead on the practice of AI-assisted development.", "reason": "Its object is the practice/process of AI-based software development itself, matching the studies/position boundary.", "source": "crawl"}, {"paper": {"id": "2607.27249", "title": "OwlPath: Lossless Knowledge Compression for LLM Bug Repair", "authors": ["Bo Zhang", "Ren Pan", "Huan Chen", "Xiang Song"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-28", "links": {"paper": "https://arxiv.org/abs/2607.27249", "github": "", "website": ""}}, "category": "software_debugging", "tags": [], "summary": "OwlPath is an OWL2 ontology-based reasoning layer atop CodeGraph that enables structural, multi-hop code retrieval to help LLM bug-repair agents resolve SWE-bench Pro issues with fewer tokens and higher recall than plain-text retrieval baselines.", "reason": "Serves issue-resolution agents by improving structural code retrieval for bug repair (software_debugging).", "source": "crawl"}, {"paper": {"id": "2607.28587", "title": "PAIChecker: Uncovering and Checking PR-Issue Misalignment in SWE-Bench-Like Benchmarks", "authors": ["Manyi Wang", "Junjielong Xu", "Pinjia He"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28587", "github": "", "website": ""}}, "category": "software_debugging", "tags": [], "summary": "PAIChecker is a multi-agent system that detects PR-Issue misalignment in SWE-bench-like benchmarks, finding 13.6% of SWE-bench Verified instances misaligned and achieving up to 92% detection accuracy across benchmarks.", "reason": "Benchmark routing: it validates and improves issue-resolution (SWE-bench style) benchmark construction, serving software_debugging.", "source": "crawl"}, {"paper": {"id": "2607.27271", "title": "RLPF: Reinforcement Learning from Performance Feedback for Code Generation", "authors": ["Huihao Jing", "Haozhe Cui", "Wenbin Hu", "Shaojin Chen", "Haochen Shi", "Changxuan Fan", "Yuxuan Liu", "Hanyu Yang", "Sirui Zhang", "Ziyi Chen", "Haoran Li", "Yangqiu Song"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27271", "github": "", "website": ""}}, "category": "software_code_generation", "tags": ["model"], "summary": "RLPF trains code-generation models with staged reinforcement learning rewards from execution progress and runtime improvement, teaching agents to produce not just correct but also efficient code, substantially improving Qwen3-32B on PerfCodeBench.", "reason": "Serves the code-generation task by optimizing for correctness and efficiency of newly produced programs.", "source": "crawl"}, {"paper": {"id": "2607.27877", "title": "An Empirical Study of Coordination Mode as the First-Class Citizen in From-Scratch Multi-Agent Coding", "authors": ["Yanyu Ren", "Yunfeng Bai", "Xizheng Wang", "Li Chen", "Dan Li"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.27877", "github": "", "website": ""}}, "category": "software_code_generation", "tags": ["benchmark", "empirical"], "summary": "MSEval is a benchmark and execution engine (LegoGent) for evaluating multi-agent teams building real-world full-stack software from scratch across 10 collaboration topologies, measuring functional success, latency, and cost. It finds coordination topology rivals model capability in shaping quality-cost-speed tradeoffs.", "reason": "Proposes a benchmark for multi-agent from-scratch application generation, so it routes to the code-generation task leaf per the benchmark-dominance rule rather than studies.", "source": "crawl"}, {"paper": {"id": "2607.28027", "title": "VISA: A Structured Description Protocol for Agent-Based Simulation Models Towards Machine Reproducibility", "authors": ["Zhou He"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28027", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "VISA is a structured, symbol-based description protocol for agent-based simulation models that enables machine-parseable specifications, consistency checking, and LLM-driven code generation to reproduce models across platforms (e.g., NetLogo to Python). It validates the approach by reproducing three independently authored models via its author-check-code-reproduce loop.", "reason": "The protocol's ultimate purpose is producing executable code that reproduces a specified simulation model, matching the artifact/code-generation leaf.", "source": "crawl"}, {"paper": {"id": "2607.28086", "title": "Distilling Answer Set Programming Theories from Large Language Models", "authors": ["Nelson Higuera Ruiz", "Markus Hofmarcher", "Claudiu Leoveanu-Condrei"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28086", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "This work studies whether LLMs, guided by a fixed agent harness with an ASP solver in the loop, can distill complete and correct Answer Set Programming theories from scratch, evaluated on VQA benchmarks (CLEVR, GQA, CLEVRER) across nine models. Frontier models reach near-100% accuracy on several benchmarks.", "reason": "An agent harness produces logic-program code (ASP theories) as the deliverable, matching code generation in the artifact branch.", "source": "crawl"}, {"paper": {"id": "2604.11270", "title": "Evaluating LLM Agents on Automated Software Analysis Tasks", "authors": ["Islem Bouzenia", "Cristian Cadar", "Michael Pradel"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2604.11270", "github": "", "website": ""}}, "category": "software_infrastructure", "tags": ["benchmark"], "summary": "Introduces AnalysisBench and AnalysisAgent to evaluate LLM agents on installing, configuring, and running third-party software analysis tools alongside target C/C++/Java projects, outperforming baseline agent architectures. Extended runs also uncover previously unknown defects in real tools.", "reason": "The task is enabling work—setting up and configuring tools/environments so analysis can run—matching software_infrastructure.", "source": "crawl"}, {"paper": {"id": "2607.27674", "title": "Can Large Language Models Resolve Real Java Merge Conflicts? An Evaluation with a Calibrated LLM-as-Judge", "authors": ["Bowen Shen"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.27674", "github": "", "website": ""}}, "category": "software_infrastructure", "tags": [], "summary": "This paper evaluates whether LLM agents can resolve real Java merge conflicts, building a generate-validate-retry solver and a calibrated LLM-as-judge metric validated against human labels. LLM resolutions match developer intent on ~55% of true conflicts, outperforming traditional tools mainly through higher coverage.", "reason": "Merge-conflict resolution is explicitly a version-control operation under software_infrastructure.", "source": "crawl"}, {"paper": {"id": "2607.28591", "title": "Change2Task: From Repository Changes to Executable Coding Agent Tasks and Environments", "authors": ["Haomin Qi", "Xingliang Wang", "Xuanqi Gao", "Baihui Sang", "Xin Zhang", "Minghua Ma", "Pengfei Gao", "Yu Kang", "Qingwei Lin", "Saravan Rajmohan", "Dongmei Zhang", "Qi Zhang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28591", "github": "", "website": ""}}, "category": "software_infrastructure", "tags": ["training-data"], "summary": "Change2Task converts merged pull requests from repository history into verified, executable coding-agent tasks and environments spanning bug fixes, feature addition, test generation, API migration, and security repair, reducing environment setup effort for agent training and evaluation.", "reason": "A resource paper whose contribution is constructing executable tasks and environments (enabling work) to train/evaluate coding agents across multiple lifecycle activities, matching software_infrastructure's enabling-work scope.", "source": "crawl"}, {"paper": {"id": "2607.27409", "title": "SWE-NFI: Studying and Benchmarking Coding Agents for Non-Functional Improvements", "authors": ["Pengyu Xue", "He Yang Yuan", "Xin Wang", "Junkai Chen", "Haonan Zhang", "Boyuan Chen", "Zishuo Ding", "Zhenhao Li", "Weiyi Shang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27409", "github": "", "website": ""}}, "category": "software_maintenance", "tags": ["benchmark"], "summary": "SWE-NFI benchmarks coding agents on behavior-preserving non-functional improvements (e.g., structural code quality) using 188 tasks from real merged PRs, evaluated with 92 executable rules. State-of-the-art agents lag human developers substantially on these non-functional improvement tasks.", "reason": "Evaluates agents on behavior-preserving code improvements, matching software_maintenance's refactoring/evolution scope.", "source": "crawl"}, {"paper": {"id": "2503.15669", "title": "ECO: An LLM-Driven Efficient Code Optimizer for Warehouse Scale Computers", "authors": ["Hannah Lin", "Martin Maas", "Maximilian Roquemore", "Arman Hasanzadeh", "Fred Lewis", "Yusuf Simonson", "Tzu-Wei Yang", "Amir Yazdanbakhsh", "Deniz Altinbüken", "Florin Papa", "Maggie Nolan Edmonds", "Aditya Patil", "Don Schwarz", "Satish Chandra", "Chris Kennelly", "Milad Hashemi", "Parthasarathy Ranganathan"], "venue": "arXiv 2025/03", "category": "", "published": "2025-03-19", "links": {"paper": "https://arxiv.org/abs/2503.15669", "github": "", "website": ""}}, "category": "software_maintenance", "tags": [], "summary": "ECO is an LLM-driven system deployed at Google that mines historical commits for performance anti-patterns, then automatically refactors matching code segments across a billion-line codebase, verifies, reviews, and measures production impact.", "reason": "Behavior-preserving performance optimization of existing code across a large codebase fits software_maintenance.", "source": "crawl"}, {"paper": {"id": "2607.27228", "title": "AI-assisted pre-review of open-source software submissions: an experience report from BOSC 2026", "authors": ["Tazro Ohta", "Nomi L. Harris", "Seth Carbon"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-14", "links": {"paper": "https://arxiv.org/abs/2607.27228", "github": "", "website": ""}}, "category": "software_review", "tags": [], "summary": "Describes bosc-pre-review, an agentic skill paired with Runabilly (which builds/tests submitted projects in Docker) to pre-assess open-source software submissions against a conference review rubric, with humans making final decisions.", "reason": "The agent judges submitted code/projects for conference review, matching software_review's patch/change-quality judgment scope.", "source": "crawl"}, {"paper": {"id": "2607.27528", "title": "ThreatForest: Multi-Agent Attack Tree Generation with Pluggable TTP Framework Mapping", "authors": ["Cristian Leo", "Anton Dykyi", "Danny Cortegaca", "Daniel Begimher", "Prakash Jha"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27528", "github": "", "website": ""}}, "category": "software_security", "tags": [], "summary": "Presents ThreatForest, a multi-agent pipeline that analyzes source code repositories to generate structured attack trees, maps steps to MITRE ATT&CK/CAPEC TTPs, and synthesizes mitigations.", "reason": "Repository auditing for threat modeling and mitigation generation is security-assurance work on code, matching software_security's repository-auditing scope.", "source": "crawl"}, {"paper": {"id": "2607.27359", "title": "PROGRESS: Property-Guided Regression Search for Semantic Falsification", "authors": ["Davis Tocheuk Mo", "Noshin Ulfat", "Matthew B. Dwyer", "Soneya Binta Hossain"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27359", "github": "", "website": ""}}, "category": "software_testing", "tags": [], "summary": "PROGRESS combines LLM-generated intent-driven properties with coverage-guided evolutionary regression-test search to expose semantic bugs missed by regression-derived oracles. Evaluated on 25 Java systems, it detects far more real bugs than regression-test generation or standalone property-based testing.", "reason": "Proactive test/property generation for correctness assurance independent of a reported defect fits software_testing.", "source": "crawl"}, {"paper": {"id": "2607.28271", "title": "Agentic Method for Deterministic Validation of Legacy Code Migration", "authors": ["Andras Ferenczi", "Jordan Docherty", "Mariya Bessonov", "Matthew Findlay", "Krishna Lingamneni"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28271", "github": "", "website": ""}}, "category": "software_testing", "tags": [], "summary": "Proposes 'Locksmith Loop,' an agentic test-synthesis method that runs instrumented COBOL and migrated Java side-by-side, iteratively searching inputs and mutations to validate migration correctness via coverage and deterministic parity checks. Evaluated on three COBOL-to-Java case studies achieving near-complete branch coverage.", "reason": "The task is proactive test/witness generation to validate correctness of migrated code, matching software_testing's coverage and verification focus.", "source": "crawl"}, {"paper": {"id": "2607.27235", "title": "RadHarmony: Radiological Data Handling in the Era of Agentic AI", "authors": ["Frank Li", "Bardia Khosravi", "Mohammadreza Chavoshi", "Theo Dapamede", "YoungSeok Jeon", "Janice Newsome", "Hari Trivedi", "Judy Gichoya"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.27235", "github": "", "website": ""}}, "category": "software_development", "tags": [], "summary": "RadHarmony is a Python library unifying heterogeneous radiological datasets, including an AI-agent skill that guides integrating new datasets into the library end-to-end, from raw data inspection through code generation and testing.", "reason": "The agent implements new dataset-integration functionality inside an existing codebase, matching feature development into an existing repository.", "source": "crawl"}, {"paper": {"id": "2607.27380", "title": "VideoCoCo: Code-as-CoT for Physically-Consistent Video Generation via an Agentic Dual-Engine System", "authors": ["Haodong Li", "Tianfei Ren", "Xiaoxiao Ma", "Chunmei Qing", "Zhen Fang", "Sipeng He", "Ziyu Guo", "Haoyu Wu", "Juanxi Tian", "Yihang Zou", "Ruichuan An", "Dongzhi Jiang", "Boxue Yang", "Ji Xie", "Xu Huang", "Wenhao Yan", "Jialv Zou", "Zhengrong Yue", "Yaxin Luo", "Xiaotong Li", "Yuzhu Wang", "Junyan Ye", "Jinjing Zhao", "Zehui Chen", "Lin Chen", "Renye Yan", "Feng Zhao", "Pheng-Ann Heng"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-29", "links": {"paper": "https://arxiv.org/abs/2607.27380", "github": "", "website": ""}}, "category": "cad", "tags": ["training-data"], "summary": "VideoCoCo uses a coding agent to synthesize executable Blender programs that construct a scene's spatiotemporal evolution as a process-level chain of thought, which is rendered into a draft and then refined into photorealistic video. A curated dataset (VideoCoCo-3K) is released to adapt the video editor to these simulated drafts.", "reason": "A coding agent produces executable 3D-scene code (Blender program) as the mechanism, matching CAD's 3D-scene-construction-as-code inclusion.", "source": "crawl"}, {"paper": {"id": "2607.27558", "title": "Drawing-Recode: Annotation Grounding for Parametric CAD Code Generation from Raster 2D CAD Drawings", "authors": ["Mingi Kim", "Yongjun Kim", "Hyungki Kim"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.27558", "github": "", "website": ""}}, "category": "cad", "tags": [], "summary": "Drawing-Recode recovers parametric CAD sequences as structured CAD code from raster 2D CAD drawings, grounding dimensional annotations to geometry via cross-attention and a new Annotation Grounding Loss. An LLM generates the final CAD code, improving over baselines even on scanned industrial drawings.", "reason": "Delivers CAD programs as code from drawings, matching the cad leaf's definition of producing 3D/CAD code artifacts.", "source": "crawl"}, {"paper": {"id": "2607.28050", "title": "IndustryForge-27B: A Domain-Enhanced Multimodal Foundation Model for Industrial CAD", "authors": ["Nianchen Deng", "Jiaxin Ai", "Tao Hu", "Shu Zou", "Yurui Dong", "Siqi Li", "Xinyu Cai", "Xuemeng Yang", "Licheng Wen", "Hongbin Zhou", "Hairong Zhang", "Pinlong Cai", "Botian Shi"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28050", "github": "", "website": ""}}, "category": "cad", "tags": ["model"], "summary": "IndustryForge-27B is a multimodal foundation model fine-tuned on industrial CAD corpora to generate parametric CAD code and COM API scripts from engineering drawings and geometry. It outperforms GPT-5.4 on CAD benchmarks while retaining general capability.", "reason": "Specialized model for producing CAD/parametric code, not a general-purpose flagship model, so routes to the cad artifact leaf plus model tag.", "source": "crawl"}, {"paper": {"id": "2607.28033", "title": "DataClawEval: A Benchmark for Data Engineering Agents in Real Industrial Harness", "authors": ["Debin Meng", "Jiaming Yang", "Zefang Zong", "Tengyue Xu", "Haining Xie", "Yang Li", "Peng Chen"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.28033", "github": "", "website": ""}}, "category": "database", "tags": ["benchmark"], "summary": "DataClawEval is a benchmark of 100 production-grade data engineering tasks spanning PySpark, MySQL, HiveSQL, Presto/Trino and FlinkSQL, graded deterministically in sandboxed environments. It evaluates 16 frontier agents, showing best performance of only 74.9 and strong domain specialization across engines.", "reason": "Tasks center on writing/executing SQL and data-pipeline code across engines, closest to the database artifact leaf.", "source": "crawl"}, {"paper": {"id": "2607.27699", "title": "RefineSVG: Visual Feedback-Driven Reinforcement Learning for Image-to-SVG Generation", "authors": ["Shaobo Liu", "Feiqiao Mao", "Shuaishuai Zhou", "Yan Zhan", "Weiqi Tan", "Zhiqiong Lu", "Zhengping Liang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-30", "links": {"paper": "https://arxiv.org/abs/2607.27699", "github": "", "website": ""}}, "category": "graphics", "tags": [], "summary": "RefineSVG is a closed-loop visual-feedback framework where an MLLM generates SVG code, renders it, compares against the target image via a visual residual map, and iteratively self-corrects using RL-trained agentic refinement. It improves reconstruction fidelity and code efficiency over open-loop SVG generation.", "reason": "The deliverable is vector-graphic (SVG) code produced from an image spec, matching artifact/graphics.", "source": "crawl"}]