[{"paper": {"id": "2607.21652", "title": "Vibe Coding in Software Development: A Multivocal Literature Review", "authors": ["Shahbaz Siddeeq", "Muhammad Waseem", "Kai-Kristian Kemell", "Mika Saari", "Jussi Rasku", "Pekka Abrahamsson"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-22", "links": {"paper": "https://arxiv.org/abs/2607.21652", "github": "", "website": ""}}, "category": "studies", "tags": ["survey"], "summary": "A multivocal literature review synthesizing 47 peer-reviewed and grey-literature sources on 'vibe coding' practices, defining it as an iterative generation-evaluation-revision loop rather than one-shot prompting. It maps evidence on productivity gains, governance, and gaps in maintainability and safety-critical use.", "reason": "Field-wide survey whose object is the vibe-coding practice itself, not a proposed agent or task.", "source": "crawl"}, {"paper": {"id": "2607.21656", "title": "Cross-Model LLM Code Review: Should you use Claude to review Codex or vice versa?", "authors": ["Zuodong Xiang", "Yike Zhang", "YueMing Zhang", "Hailu Xu"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-22", "links": {"paper": "https://arxiv.org/abs/2607.21656", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "A controlled experiment testing whether pairing Claude and Codex coding agents as writer/reviewer improves LeetCode-style task success, across solo, cross-model, and same-model conditions. Results show an asymmetric benefit: Claude reviewing Codex substantially raises pass rates, while the reverse pairing hurts performance.", "reason": "An empirical study of existing agents' interaction behavior rather than a proposed review method, so it studies the practice.", "source": "crawl"}, {"paper": {"id": "2607.21674", "title": "Output Format x Model Identity: Interaction Effects in Single-Round Coding Agent Performance", "authors": ["Yang Yang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-23", "links": {"paper": "https://arxiv.org/abs/2607.21674", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "A controlled single-round experiment crossing three LLMs and three output formats (full file, JSON Patch, unified diff) across six coding tasks, finding strong model-by-format interaction effects and a 'format misuse' failure mode where correct fixes are applied with excessive scope. No universal best output format exists across models.", "reason": "An empirical behavioral study of coding-agent performance sensitivity to output format, not a proposed method or benchmark task.", "source": "crawl"}, {"paper": {"id": "2607.21832", "title": "How Do AI Coding Agents Contribute to Software Development? an Empirical Study of Agentic Pull Requests", "authors": ["Iren Mazloomzadeh", "Mohammad Mehdi Morovati", "Foutse Khomh"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-23", "links": {"paper": "https://arxiv.org/abs/2607.21832", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "Using the AIDev dataset, this empirical study characterizes AI-generated pull requests versus human PRs, analyzing merge rates, task distributions, and quality-related characteristics across software development lifecycle stages over time. It provides a longitudinal empirical perspective on coding agents' real-world contributions.", "reason": "An empirical/behavioral study whose object is coding agents' practice across the development lifecycle, matching the studies leaf.", "source": "crawl"}, {"paper": {"id": "2607.21997", "title": "\"Go Home Copilot, You're Drunk\": Understanding Developer Responses to Agent-Generated Code Review Comments", "authors": ["Shamse Tasnim Cynthia", "Ratnadira Widyasari", "Banani Roy", "Ting Zhang", "David Lo"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.21997", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "A large-scale empirical study analyzing 54,791 agent-generated code review comments from five AI coding agents across 342 GitHub repositories, examining resolution rates, developer roles, and factors predicting comment usefulness. Identifies ten discussion patterns explaining unresolved feedback.", "reason": "An empirical/behavioral study of human-agent interaction around code review, not a proposed method, fits studies.", "source": "crawl"}, {"paper": {"id": "2607.22368", "title": "Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic AI", "authors": ["Jiaqi Shao", "Hanck Chen", "Wei Zhang", "Maxm Pan", "Bing Luo"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22368", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "Introduces protocol validity and HackDetect, a post-hoc audit method that quantifies reward hacking and score inflation across 15 agent benchmarks spanning repository editing, web research, and terminal use, finding widespread exploits.", "reason": "The paper's object of study is agent benchmarks and their evaluation validity across many activities, making it a field-wide empirical study rather than a task-solving agent.", "source": "crawl"}, {"paper": {"id": "2607.22406", "title": "Vibe Coding: An Experiment with Test-Driven Development", "authors": ["Moritz Mock", "Barbara Russo"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22406", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "An exploratory empirical study comparing solo, human-CLLM collaborative, fully automated, and agentic (MetaGPT-X) workflows for Test-Driven Development, examining code and test quality outcomes. Findings show agentic workflows favor rapid, functionally correct code while collaborative workflows yield better-organized test suites.", "reason": "The paper's object is studying human-AI collaboration patterns in coding practice, not proposing an agent for a specific task, so it fits the empirical studies leaf.", "source": "crawl"}, {"paper": {"id": "2607.22471", "title": "MineValiCoder: Reliable Code Generation with Test Case Quality Mining and Bipartite Graph-Based Mutual Validation", "authors": ["Zhen Zhao", "Qihang Yang", "Feifei Dai", "Xiangfang Li", "Bo Li"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22471", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "MineValiCoder is a closed-loop test-driven-development framework that mines reliable test cases and uses bipartite-graph mutual validation between generated code and tests to select optimal code candidates from natural-language requirements. It improves Pass@1 over prior methods on HumanEval, MBPP, APPS, and LiveCodeBench.", "reason": "The task's ultimate purpose is producing code from NL specs via an iterative generate-test-validate loop, matching software_code_generation.", "source": "crawl"}, {"paper": {"id": "2607.21618", "title": "LeafData: An Agentic System for Data Migration", "authors": ["Sadanand Katukuri", "Rajasekhar Bada", "Navya Induri", "Rohit Gandham", "Lynette Pinto", "Joses Selvan", "Abishek Krishnamoorthy", "Joseph Rozario", "Pu Tian", "Pavan Poudel", "Yalong Wu"], "venue": "arXiv 2026/05", "category": "", "published": "2026-05-31", "links": {"paper": "https://arxiv.org/abs/2607.21618", "github": "", "website": ""}}, "category": "software_infrastructure", "tags": [], "summary": "LeafData is an agentic system with a chatbot frontend and backend that converts user intent into validated, executable JSON pipeline configuration for data migration across heterogeneous sources, enabling orchestration without manual coding.", "reason": "The generated configuration is enabling/deployment artifact code that lets migration pipelines run, matching the environment-setup/CI-CD enabling-work leaf.", "source": "crawl"}, {"paper": {"id": "2607.21909", "title": "Claim Plane: Enforceable Change Intents and Dynamic Scope for Parallel Coding Agents", "authors": ["Maxim Nikolaev"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.21909", "github": "", "website": ""}}, "category": "software_infrastructure", "tags": [], "summary": "Claim Plane is a coordination architecture that admits parallel coding-agent changes via versioned intents, deterministic scope control, and Git-tree provenance to prevent integration-time interference. It uses worktree locks, fencing tokens, and scope promotion to serialize overlapping edits safely.", "reason": "Coordinates concurrent version-control/merge operations among coding agents, matching software_infrastructure's version-control scope.", "source": "crawl"}, {"paper": {"id": "2607.21677", "title": "Enhancing SLMs for Sustainable Code Optimization in Radio-Astronomy", "authors": ["Elisa Chiarotto", "Jingbo Li", "P. Chris Broekema", "Rob V. van Nieuwpoort"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-23", "links": {"paper": "https://arxiv.org/abs/2607.21677", "github": "", "website": ""}}, "category": "software_maintenance", "tags": [], "summary": "An agentic approach using small language models with multi-sample generation and compiler-feedback loops to optimize and port large radio-astronomy (LOFAR) codebases to hardware accelerators, aiming for energy-efficient performance improvement. The method matches larger single-generation LLMs while using fewer resources.", "reason": "The served task is performance optimization and porting of existing code while preserving behavior, matching software_maintenance.", "source": "crawl"}, {"paper": {"id": "2607.22095", "title": "Code Review is a Conversation: Toward Conversational AI Review Assistants", "authors": ["Rosalia Tufano"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22095", "github": "", "website": ""}}, "category": "software_review", "tags": ["position"], "summary": "A vision paper arguing that AI code review tools should become conversational partners engaging in dialogue, rationale-transfer, and negotiation rather than one-shot comment generators. It outlines a research agenda for designing and evaluating such conversational review assistants.", "reason": "A position paper focused on the single activity of code review, so it stays in software_review with a position tag rather than the field-wide studies leaf.", "source": "crawl"}, {"paper": {"id": "2607.21641", "title": "Tool-Guided Retrieval-Augmented Repair for Securing LLM-Generated C Code", "authors": ["Vidyut Sriram", "Saatvik Pradhan", "Suman Saha"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-21", "links": {"paper": "https://arxiv.org/abs/2607.21641", "github": "", "website": ""}}, "category": "software_security", "tags": [], "summary": "Presents an analysis-and-repair workflow combining compiler diagnostics, CodeQL static analysis, KLEE symbolic execution, and retrieval of prior fixes to iteratively reduce compilation failures and security defects in LLM-generated C code.", "reason": "The served task is hardening/repairing generated code against vulnerabilities, matching the software_security leaf.", "source": "crawl"}, {"paper": {"id": "2607.21983", "title": "Ethereum NFT Smart Contracts: Knowledge-Guided Vulnerability Detection with LLM and Code Slicing", "authors": ["Deyu Yang", "Rundong Wei", "Xiaoqi Li"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.21983", "github": "", "website": ""}}, "category": "software_security", "tags": [], "summary": "Presents an LLM-based vulnerability detection pipeline for Ethereum NFT smart contracts, combining regex-based code slicing, an ERC-721 knowledge base, and constrained DeepSeek analysis. Evaluated on 450 contracts, it improves detection rate over ablated variants lacking domain knowledge or slicing.", "reason": "Automated vulnerability detection/auditing of smart contract code fits software_security.", "source": "crawl"}, {"paper": {"id": "2607.22076", "title": "PoCEvolve: Generating Proof-of-Concept Exploits from Security Patches with Vulnerability-Aware Prompt Evolution", "authors": ["Duc Manh Tran", "Ratnadira Widyasari", "Ivana Clairine Irsan", "Huihui Huang", "Ting Zhang", "Shar Lwin Khin", "Ouh Eng Lieh", "Hong Jin Kang", "David Lo"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22076", "github": "", "website": ""}}, "category": "software_security", "tags": [], "summary": "PoCEvolve is a vulnerability-aware prompt-evolution framework that automatically synthesizes proof-of-concept exploits directly from security-fixing commits, without needing a detailed vulnerability report. It substantially outperforms prior PoC-generation baselines on JavaScript CVE benchmarks.", "reason": "Generating exploit PoCs from patches to validate and expose vulnerabilities is a security-assurance activity on code.", "source": "crawl"}, {"paper": {"id": "2606.17981", "title": "Planning to Hammer: Difficulty-Aware Decomposition for Automating Rocq Proofs", "authors": ["Ning Zhang", "Nongyu Di", "Zenan Li", "Yuan Yao", "Xiaoxing Ma"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2606.17981", "github": "", "website": ""}}, "category": "software_testing", "tags": [], "summary": "Quarry is a planning-based proof synthesis framework for Rocq that separates LLM-driven proof decomposition from automated tactic execution (CoqHammer), recursively solving sublemmas to improve proof automation for software correctness verification.", "reason": "An agent that writes and executes formal proof code, framed around ensuring software correctness, fits artifact/software_testing (formal verification of software).", "source": "crawl"}, {"paper": {"id": "2607.21957", "title": "KaPilot: LLM-Assisted Generation of Kani Specifications for Unsafe Rust Verification", "authors": ["Minghua Wang", "Yuxi Ling", "Mingzhi Gao", "Yuwei Liu", "Lin Huang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.21957", "github": "", "website": ""}}, "category": "software_testing", "tags": [], "summary": "KaPilot is a multi-agent framework that generates and iteratively refines Kani specifications to formally verify memory safety of unsafe Rust functions. It uses safety-requirement extraction, generate-precheck-verify loops, and a selection strategy, outperforming AutoSpec on verifiable specification generation.", "reason": "Formal verification of software correctness via agentic specification generation falls under software_testing.", "source": "crawl"}, {"paper": {"id": "2607.22161", "title": "HarnessLLM: Rust Verification Harness Generation with Large Language Models", "authors": ["Minghua Wang", "Yuwei Liu", "Lin Huang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22161", "github": "", "website": ""}}, "category": "software_testing", "tags": [], "summary": "HarnessLLM automates generation of Rust verification harnesses from existing test suites using LLMs, extracting calling scenarios and synthesizing nondeterministic arguments for memory-safety formal verification. It outperforms prior tools and discovers real memory-safety bugs across nine codebases.", "reason": "Formal verification harness generation for software correctness falls under software_testing per its explicit inclusion of proof-assisted verification.", "source": "crawl"}, {"paper": {"id": "2605.08971", "title": "Extrusion Segmentation Strategy to improve CAD Reconstruction from Point Cloud", "authors": ["Said Harb", "Mehdi Maboudi", "Markus Gerke"], "venue": "arXiv 2026/05", "category": "", "published": "2026-05-09", "links": {"paper": "https://arxiv.org/abs/2605.08971", "github": "", "website": ""}}, "category": "cad", "tags": [], "summary": "Proposes an extrusion-based segmentation strategy that decomposes point clouds into partial shapes to improve deep-learning reconstruction of CAD models from 3D scans.", "reason": "The task's purpose is producing CAD-as-code models from point clouds, matching the cad artifact leaf.", "source": "crawl"}, {"paper": {"id": "2601.18119", "title": "Beyond Text-to-SQL: Can LLMs Really Debug Enterprise ETL SQL?", "authors": ["Jing Ye", "Yiwen Duan", "Yonghong Yu", "Victor Ma", "Yang Gao", "Xing Chen"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2601.18119", "github": "", "website": ""}}, "category": "database", "tags": ["benchmark"], "summary": "Introduces OurBench, a benchmark of enterprise ETL SQL queries with injected syntax and semantic bugs, evaluating nearly 30 LLMs' ability to debug complex real-world SQL code.", "reason": "The served task is producing/repairing SQL code, which routes to the database leaf per its query-synthesis and debugging scope.", "source": "crawl"}, {"paper": {"id": "2607.22115", "title": "Benchmarking Text-to-SQL under Role-Based Access Control", "authors": ["Yang Fei", "Yangfan Jiang", "Yin Yang", "Xiaokui Xiao"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22115", "github": "", "website": ""}}, "category": "database", "tags": ["benchmark"], "summary": "Introduces a benchmark that augments text-to-SQL datasets with realistic role-based access control policies and evaluates LLM systems for both SQL correctness and access-control compliance. Finds many high-scoring systems degrade sharply once RBAC constraints are enforced.", "reason": "Text-to-SQL query synthesis is the served task, matching the database leaf's explicit inclusion of text-to-SQL benchmarks.", "source": "crawl"}, {"paper": {"id": "2607.22511", "title": "CausalForge: A Formally Grounded, Self-Improving Agentic Framework for Automated Research in Causal Inference", "authors": ["Jiyuan Tan", "Vasilis Syrgkanis"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22511", "github": "", "website": ""}}, "category": "world_research", "tags": [], "summary": "CausalForge is a self-improving agentic pipeline that autonomously selects causal-inference research topics, formalizes claims, and constructs machine-checked Lean proofs, backed by a 7,035-declaration Lean library and a statement-audit step comparing formal theorems to informal claims. It evaluates the system on artifacts from completed autonomous research runs.", "reason": "A Lean proof-assistant agent whose purpose is mathematical/theoretical knowledge fits world_research's pure theorem-proving example.", "source": "crawl"}, {"paper": {"id": "2607.21642", "title": "CARE: Pre-Execution Command Verification for Shell-Executing LLM Agents", "authors": ["Wenxiao Zhang", "Yu Liu", "Zhiwei Yang", "Zhongyi Zhang", "Hanqi Feng", "Xinyu Wang", "Peng Qiu", "Yanbing Liu", "Barnabas Poczos", "Jin B. Hong"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-21", "links": {"paper": "https://arxiv.org/abs/2607.21642", "github": "", "website": ""}}, "category": "world_terminal", "tags": [], "summary": "CARE is a pre-execution verifier that canonicalizes and risk-assesses shell commands produced by LLM agents before execution, escalating only ambiguous cases to an LLM judge to reduce harmful command execution.", "reason": "Shell commands are code and the terminal is the agent's world, so mediating their execution fits world_terminal, not the excluded GUI-agent security case.", "source": "crawl"}, {"paper": {"id": "2607.21763", "title": "Every Model Cheats: Prompt-Level Mitigation of Cheating on Offensive Cyber Tasks", "authors": ["Michael Kouremetis", "Ads Dawson", "Raja Sekhar Rao Dheekonda", "Brian Greunke"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-23", "links": {"paper": "https://arxiv.org/abs/2607.21763", "github": "", "website": ""}}, "category": "world_terminal", "tags": ["empirical"], "summary": "A large-scale audited study of 22 frontier LLM agents on Cybench CTF challenges finds pervasive cheating (up to 37% of passes), and shows prompt-level anti-cheat instructions substantially reduce (but do not eliminate) cheating without hurting solve rates. The authors propose a 'solve rate' metric to separate genuine capability from cheated outcomes.", "reason": "Empirical/behavioral study of agents acting via shell/infrastructure on offensive-security CTF tasks, a single-world study routed to world_terminal.", "source": "crawl"}, {"paper": {"id": "2607.22031", "title": "IDSTune: A Multi-Agent Collaborative Framework for Integrated Database System Tuning", "authors": ["Yiyan Li", "Guanli Liu", "Renata Borovica-Gajic", "Haoyang Li", "Zihang Qiu", "Xinmei Huang", "Andreas Kipf", "Cuiping Li", "Hong Chen"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22031", "github": "", "website": ""}}, "category": "world_terminal", "tags": [], "summary": "IDSTune is an LLM-driven multi-agent framework that jointly tunes database knobs, indexes, and materialized views by compressing workloads and coordinating specialized tuning agents. It achieves large performance gains and faster tuning across diverse workloads.", "reason": "Agents operate and tune a running database system, which the taxonomy routes to world_terminal rather than the database artifact leaf.", "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 92 out-of-scope; 0 failed (retried next run).
1. Vibe Coding in Software Development: A Multivocal Literature Review
Shahbaz Siddeeq, Muhammad Waseem, Kai-Kristian Kemell, et al. · arXiv 2026/07 · paper
proposed:
studies· tags:survey2. Cross-Model LLM Code Review: Should you use Claude to review Codex or vice versa?
Zuodong Xiang, Yike Zhang, YueMing Zhang, et al. · arXiv 2026/07 · paper
proposed:
studies· tags:empirical3. Output Format x Model Identity: Interaction Effects in Single-Round Coding Agent Performance
Yang Yang · arXiv 2026/07 · paper
proposed:
studies· tags:empirical4. How Do AI Coding Agents Contribute to Software Development? an Empirical Study of Agentic Pull Requests
Iren Mazloomzadeh, Mohammad Mehdi Morovati, Foutse Khomh · arXiv 2026/07 · paper
proposed:
studies· tags:empirical5. "Go Home Copilot, You're Drunk": Understanding Developer Responses to Agent-Generated Code Review Comments
Shamse Tasnim Cynthia, Ratnadira Widyasari, Banani Roy, et al. · arXiv 2026/07 · paper
proposed:
studies· tags:empirical6. Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic AI
Jiaqi Shao, Hanck Chen, Wei Zhang, et al. · arXiv 2026/07 · paper
proposed:
studies· tags:empirical7. Vibe Coding: An Experiment with Test-Driven Development
Moritz Mock, Barbara Russo · arXiv 2026/07 · paper
proposed:
studies· tags:empirical8. MineValiCoder: Reliable Code Generation with Test Case Quality Mining and Bipartite Graph-Based Mutual Validation
Zhen Zhao, Qihang Yang, Feifei Dai, et al. · arXiv 2026/07 · paper
proposed:
software_code_generation· tags: none9. LeafData: An Agentic System for Data Migration
Sadanand Katukuri, Rajasekhar Bada, Navya Induri, et al. · arXiv 2026/05 · paper
proposed:
software_infrastructure· tags: none10. Claim Plane: Enforceable Change Intents and Dynamic Scope for Parallel Coding Agents
Maxim Nikolaev · arXiv 2026/07 · paper
proposed:
software_infrastructure· tags: none11. Enhancing SLMs for Sustainable Code Optimization in Radio-Astronomy
Elisa Chiarotto, Jingbo Li, P. Chris Broekema, et al. · arXiv 2026/07 · paper
proposed:
software_maintenance· tags: none12. Code Review is a Conversation: Toward Conversational AI Review Assistants
Rosalia Tufano · arXiv 2026/07 · paper
proposed:
software_review· tags:position13. Tool-Guided Retrieval-Augmented Repair for Securing LLM-Generated C Code
Vidyut Sriram, Saatvik Pradhan, Suman Saha · arXiv 2026/07 · paper
proposed:
software_security· tags: none14. Ethereum NFT Smart Contracts: Knowledge-Guided Vulnerability Detection with LLM and Code Slicing
Deyu Yang, Rundong Wei, Xiaoqi Li · arXiv 2026/07 · paper
proposed:
software_security· tags: none15. PoCEvolve: Generating Proof-of-Concept Exploits from Security Patches with Vulnerability-Aware Prompt Evolution
Duc Manh Tran, Ratnadira Widyasari, Ivana Clairine Irsan, et al. · arXiv 2026/07 · paper
proposed:
software_security· tags: none16. Planning to Hammer: Difficulty-Aware Decomposition for Automating Rocq Proofs
Ning Zhang, Nongyu Di, Zenan Li, et al. · arXiv 2026/07 · paper
proposed:
software_testing· tags: none17. KaPilot: LLM-Assisted Generation of Kani Specifications for Unsafe Rust Verification
Minghua Wang, Yuxi Ling, Mingzhi Gao, et al. · arXiv 2026/07 · paper
proposed:
software_testing· tags: none18. HarnessLLM: Rust Verification Harness Generation with Large Language Models
Minghua Wang, Yuwei Liu, Lin Huang · arXiv 2026/07 · paper
proposed:
software_testing· tags: none19. Extrusion Segmentation Strategy to improve CAD Reconstruction from Point Cloud
Said Harb, Mehdi Maboudi, Markus Gerke · arXiv 2026/05 · paper
proposed:
cad· tags: none20. Beyond Text-to-SQL: Can LLMs Really Debug Enterprise ETL SQL?
Jing Ye, Yiwen Duan, Yonghong Yu, et al. · arXiv 2026/07 · paper
proposed:
database· tags:benchmark21. Benchmarking Text-to-SQL under Role-Based Access Control
Yang Fei, Yangfan Jiang, Yin Yang, et al. · arXiv 2026/07 · paper
proposed:
database· tags:benchmark22. CausalForge: A Formally Grounded, Self-Improving Agentic Framework for Automated Research in Causal Inference
Jiyuan Tan, Vasilis Syrgkanis · arXiv 2026/07 · paper
proposed:
world_research· tags: none23. CARE: Pre-Execution Command Verification for Shell-Executing LLM Agents
Wenxiao Zhang, Yu Liu, Zhiwei Yang, et al. · arXiv 2026/07 · paper
proposed:
world_terminal· tags: none24. Every Model Cheats: Prompt-Level Mitigation of Cheating on Offensive Cyber Tasks
Michael Kouremetis, Ads Dawson, Raja Sekhar Rao Dheekonda, et al. · arXiv 2026/07 · paper
proposed:
world_terminal· tags:empirical25. IDSTune: A Multi-Agent Collaborative Framework for Integrated Database System Tuning
Yiyan Li, Guanli Liu, Renata Borovica-Gajic, et al. · arXiv 2026/07 · paper
proposed:
world_terminal· tags: nonemachine payload (do not edit)
[{"paper": {"id": "2607.21652", "title": "Vibe Coding in Software Development: A Multivocal Literature Review", "authors": ["Shahbaz Siddeeq", "Muhammad Waseem", "Kai-Kristian Kemell", "Mika Saari", "Jussi Rasku", "Pekka Abrahamsson"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-22", "links": {"paper": "https://arxiv.org/abs/2607.21652", "github": "", "website": ""}}, "category": "studies", "tags": ["survey"], "summary": "A multivocal literature review synthesizing 47 peer-reviewed and grey-literature sources on 'vibe coding' practices, defining it as an iterative generation-evaluation-revision loop rather than one-shot prompting. It maps evidence on productivity gains, governance, and gaps in maintainability and safety-critical use.", "reason": "Field-wide survey whose object is the vibe-coding practice itself, not a proposed agent or task.", "source": "crawl"}, {"paper": {"id": "2607.21656", "title": "Cross-Model LLM Code Review: Should you use Claude to review Codex or vice versa?", "authors": ["Zuodong Xiang", "Yike Zhang", "YueMing Zhang", "Hailu Xu"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-22", "links": {"paper": "https://arxiv.org/abs/2607.21656", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "A controlled experiment testing whether pairing Claude and Codex coding agents as writer/reviewer improves LeetCode-style task success, across solo, cross-model, and same-model conditions. Results show an asymmetric benefit: Claude reviewing Codex substantially raises pass rates, while the reverse pairing hurts performance.", "reason": "An empirical study of existing agents' interaction behavior rather than a proposed review method, so it studies the practice.", "source": "crawl"}, {"paper": {"id": "2607.21674", "title": "Output Format x Model Identity: Interaction Effects in Single-Round Coding Agent Performance", "authors": ["Yang Yang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-23", "links": {"paper": "https://arxiv.org/abs/2607.21674", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "A controlled single-round experiment crossing three LLMs and three output formats (full file, JSON Patch, unified diff) across six coding tasks, finding strong model-by-format interaction effects and a 'format misuse' failure mode where correct fixes are applied with excessive scope. No universal best output format exists across models.", "reason": "An empirical behavioral study of coding-agent performance sensitivity to output format, not a proposed method or benchmark task.", "source": "crawl"}, {"paper": {"id": "2607.21832", "title": "How Do AI Coding Agents Contribute to Software Development? an Empirical Study of Agentic Pull Requests", "authors": ["Iren Mazloomzadeh", "Mohammad Mehdi Morovati", "Foutse Khomh"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-23", "links": {"paper": "https://arxiv.org/abs/2607.21832", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "Using the AIDev dataset, this empirical study characterizes AI-generated pull requests versus human PRs, analyzing merge rates, task distributions, and quality-related characteristics across software development lifecycle stages over time. It provides a longitudinal empirical perspective on coding agents' real-world contributions.", "reason": "An empirical/behavioral study whose object is coding agents' practice across the development lifecycle, matching the studies leaf.", "source": "crawl"}, {"paper": {"id": "2607.21997", "title": "\"Go Home Copilot, You're Drunk\": Understanding Developer Responses to Agent-Generated Code Review Comments", "authors": ["Shamse Tasnim Cynthia", "Ratnadira Widyasari", "Banani Roy", "Ting Zhang", "David Lo"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.21997", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "A large-scale empirical study analyzing 54,791 agent-generated code review comments from five AI coding agents across 342 GitHub repositories, examining resolution rates, developer roles, and factors predicting comment usefulness. Identifies ten discussion patterns explaining unresolved feedback.", "reason": "An empirical/behavioral study of human-agent interaction around code review, not a proposed method, fits studies.", "source": "crawl"}, {"paper": {"id": "2607.22368", "title": "Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic AI", "authors": ["Jiaqi Shao", "Hanck Chen", "Wei Zhang", "Maxm Pan", "Bing Luo"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22368", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "Introduces protocol validity and HackDetect, a post-hoc audit method that quantifies reward hacking and score inflation across 15 agent benchmarks spanning repository editing, web research, and terminal use, finding widespread exploits.", "reason": "The paper's object of study is agent benchmarks and their evaluation validity across many activities, making it a field-wide empirical study rather than a task-solving agent.", "source": "crawl"}, {"paper": {"id": "2607.22406", "title": "Vibe Coding: An Experiment with Test-Driven Development", "authors": ["Moritz Mock", "Barbara Russo"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22406", "github": "", "website": ""}}, "category": "studies", "tags": ["empirical"], "summary": "An exploratory empirical study comparing solo, human-CLLM collaborative, fully automated, and agentic (MetaGPT-X) workflows for Test-Driven Development, examining code and test quality outcomes. Findings show agentic workflows favor rapid, functionally correct code while collaborative workflows yield better-organized test suites.", "reason": "The paper's object is studying human-AI collaboration patterns in coding practice, not proposing an agent for a specific task, so it fits the empirical studies leaf.", "source": "crawl"}, {"paper": {"id": "2607.22471", "title": "MineValiCoder: Reliable Code Generation with Test Case Quality Mining and Bipartite Graph-Based Mutual Validation", "authors": ["Zhen Zhao", "Qihang Yang", "Feifei Dai", "Xiangfang Li", "Bo Li"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22471", "github": "", "website": ""}}, "category": "software_code_generation", "tags": [], "summary": "MineValiCoder is a closed-loop test-driven-development framework that mines reliable test cases and uses bipartite-graph mutual validation between generated code and tests to select optimal code candidates from natural-language requirements. It improves Pass@1 over prior methods on HumanEval, MBPP, APPS, and LiveCodeBench.", "reason": "The task's ultimate purpose is producing code from NL specs via an iterative generate-test-validate loop, matching software_code_generation.", "source": "crawl"}, {"paper": {"id": "2607.21618", "title": "LeafData: An Agentic System for Data Migration", "authors": ["Sadanand Katukuri", "Rajasekhar Bada", "Navya Induri", "Rohit Gandham", "Lynette Pinto", "Joses Selvan", "Abishek Krishnamoorthy", "Joseph Rozario", "Pu Tian", "Pavan Poudel", "Yalong Wu"], "venue": "arXiv 2026/05", "category": "", "published": "2026-05-31", "links": {"paper": "https://arxiv.org/abs/2607.21618", "github": "", "website": ""}}, "category": "software_infrastructure", "tags": [], "summary": "LeafData is an agentic system with a chatbot frontend and backend that converts user intent into validated, executable JSON pipeline configuration for data migration across heterogeneous sources, enabling orchestration without manual coding.", "reason": "The generated configuration is enabling/deployment artifact code that lets migration pipelines run, matching the environment-setup/CI-CD enabling-work leaf.", "source": "crawl"}, {"paper": {"id": "2607.21909", "title": "Claim Plane: Enforceable Change Intents and Dynamic Scope for Parallel Coding Agents", "authors": ["Maxim Nikolaev"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.21909", "github": "", "website": ""}}, "category": "software_infrastructure", "tags": [], "summary": "Claim Plane is a coordination architecture that admits parallel coding-agent changes via versioned intents, deterministic scope control, and Git-tree provenance to prevent integration-time interference. It uses worktree locks, fencing tokens, and scope promotion to serialize overlapping edits safely.", "reason": "Coordinates concurrent version-control/merge operations among coding agents, matching software_infrastructure's version-control scope.", "source": "crawl"}, {"paper": {"id": "2607.21677", "title": "Enhancing SLMs for Sustainable Code Optimization in Radio-Astronomy", "authors": ["Elisa Chiarotto", "Jingbo Li", "P. Chris Broekema", "Rob V. van Nieuwpoort"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-23", "links": {"paper": "https://arxiv.org/abs/2607.21677", "github": "", "website": ""}}, "category": "software_maintenance", "tags": [], "summary": "An agentic approach using small language models with multi-sample generation and compiler-feedback loops to optimize and port large radio-astronomy (LOFAR) codebases to hardware accelerators, aiming for energy-efficient performance improvement. The method matches larger single-generation LLMs while using fewer resources.", "reason": "The served task is performance optimization and porting of existing code while preserving behavior, matching software_maintenance.", "source": "crawl"}, {"paper": {"id": "2607.22095", "title": "Code Review is a Conversation: Toward Conversational AI Review Assistants", "authors": ["Rosalia Tufano"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22095", "github": "", "website": ""}}, "category": "software_review", "tags": ["position"], "summary": "A vision paper arguing that AI code review tools should become conversational partners engaging in dialogue, rationale-transfer, and negotiation rather than one-shot comment generators. It outlines a research agenda for designing and evaluating such conversational review assistants.", "reason": "A position paper focused on the single activity of code review, so it stays in software_review with a position tag rather than the field-wide studies leaf.", "source": "crawl"}, {"paper": {"id": "2607.21641", "title": "Tool-Guided Retrieval-Augmented Repair for Securing LLM-Generated C Code", "authors": ["Vidyut Sriram", "Saatvik Pradhan", "Suman Saha"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-21", "links": {"paper": "https://arxiv.org/abs/2607.21641", "github": "", "website": ""}}, "category": "software_security", "tags": [], "summary": "Presents an analysis-and-repair workflow combining compiler diagnostics, CodeQL static analysis, KLEE symbolic execution, and retrieval of prior fixes to iteratively reduce compilation failures and security defects in LLM-generated C code.", "reason": "The served task is hardening/repairing generated code against vulnerabilities, matching the software_security leaf.", "source": "crawl"}, {"paper": {"id": "2607.21983", "title": "Ethereum NFT Smart Contracts: Knowledge-Guided Vulnerability Detection with LLM and Code Slicing", "authors": ["Deyu Yang", "Rundong Wei", "Xiaoqi Li"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.21983", "github": "", "website": ""}}, "category": "software_security", "tags": [], "summary": "Presents an LLM-based vulnerability detection pipeline for Ethereum NFT smart contracts, combining regex-based code slicing, an ERC-721 knowledge base, and constrained DeepSeek analysis. Evaluated on 450 contracts, it improves detection rate over ablated variants lacking domain knowledge or slicing.", "reason": "Automated vulnerability detection/auditing of smart contract code fits software_security.", "source": "crawl"}, {"paper": {"id": "2607.22076", "title": "PoCEvolve: Generating Proof-of-Concept Exploits from Security Patches with Vulnerability-Aware Prompt Evolution", "authors": ["Duc Manh Tran", "Ratnadira Widyasari", "Ivana Clairine Irsan", "Huihui Huang", "Ting Zhang", "Shar Lwin Khin", "Ouh Eng Lieh", "Hong Jin Kang", "David Lo"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22076", "github": "", "website": ""}}, "category": "software_security", "tags": [], "summary": "PoCEvolve is a vulnerability-aware prompt-evolution framework that automatically synthesizes proof-of-concept exploits directly from security-fixing commits, without needing a detailed vulnerability report. It substantially outperforms prior PoC-generation baselines on JavaScript CVE benchmarks.", "reason": "Generating exploit PoCs from patches to validate and expose vulnerabilities is a security-assurance activity on code.", "source": "crawl"}, {"paper": {"id": "2606.17981", "title": "Planning to Hammer: Difficulty-Aware Decomposition for Automating Rocq Proofs", "authors": ["Ning Zhang", "Nongyu Di", "Zenan Li", "Yuan Yao", "Xiaoxing Ma"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2606.17981", "github": "", "website": ""}}, "category": "software_testing", "tags": [], "summary": "Quarry is a planning-based proof synthesis framework for Rocq that separates LLM-driven proof decomposition from automated tactic execution (CoqHammer), recursively solving sublemmas to improve proof automation for software correctness verification.", "reason": "An agent that writes and executes formal proof code, framed around ensuring software correctness, fits artifact/software_testing (formal verification of software).", "source": "crawl"}, {"paper": {"id": "2607.21957", "title": "KaPilot: LLM-Assisted Generation of Kani Specifications for Unsafe Rust Verification", "authors": ["Minghua Wang", "Yuxi Ling", "Mingzhi Gao", "Yuwei Liu", "Lin Huang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.21957", "github": "", "website": ""}}, "category": "software_testing", "tags": [], "summary": "KaPilot is a multi-agent framework that generates and iteratively refines Kani specifications to formally verify memory safety of unsafe Rust functions. It uses safety-requirement extraction, generate-precheck-verify loops, and a selection strategy, outperforming AutoSpec on verifiable specification generation.", "reason": "Formal verification of software correctness via agentic specification generation falls under software_testing.", "source": "crawl"}, {"paper": {"id": "2607.22161", "title": "HarnessLLM: Rust Verification Harness Generation with Large Language Models", "authors": ["Minghua Wang", "Yuwei Liu", "Lin Huang"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22161", "github": "", "website": ""}}, "category": "software_testing", "tags": [], "summary": "HarnessLLM automates generation of Rust verification harnesses from existing test suites using LLMs, extracting calling scenarios and synthesizing nondeterministic arguments for memory-safety formal verification. It outperforms prior tools and discovers real memory-safety bugs across nine codebases.", "reason": "Formal verification harness generation for software correctness falls under software_testing per its explicit inclusion of proof-assisted verification.", "source": "crawl"}, {"paper": {"id": "2605.08971", "title": "Extrusion Segmentation Strategy to improve CAD Reconstruction from Point Cloud", "authors": ["Said Harb", "Mehdi Maboudi", "Markus Gerke"], "venue": "arXiv 2026/05", "category": "", "published": "2026-05-09", "links": {"paper": "https://arxiv.org/abs/2605.08971", "github": "", "website": ""}}, "category": "cad", "tags": [], "summary": "Proposes an extrusion-based segmentation strategy that decomposes point clouds into partial shapes to improve deep-learning reconstruction of CAD models from 3D scans.", "reason": "The task's purpose is producing CAD-as-code models from point clouds, matching the cad artifact leaf.", "source": "crawl"}, {"paper": {"id": "2601.18119", "title": "Beyond Text-to-SQL: Can LLMs Really Debug Enterprise ETL SQL?", "authors": ["Jing Ye", "Yiwen Duan", "Yonghong Yu", "Victor Ma", "Yang Gao", "Xing Chen"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2601.18119", "github": "", "website": ""}}, "category": "database", "tags": ["benchmark"], "summary": "Introduces OurBench, a benchmark of enterprise ETL SQL queries with injected syntax and semantic bugs, evaluating nearly 30 LLMs' ability to debug complex real-world SQL code.", "reason": "The served task is producing/repairing SQL code, which routes to the database leaf per its query-synthesis and debugging scope.", "source": "crawl"}, {"paper": {"id": "2607.22115", "title": "Benchmarking Text-to-SQL under Role-Based Access Control", "authors": ["Yang Fei", "Yangfan Jiang", "Yin Yang", "Xiaokui Xiao"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22115", "github": "", "website": ""}}, "category": "database", "tags": ["benchmark"], "summary": "Introduces a benchmark that augments text-to-SQL datasets with realistic role-based access control policies and evaluates LLM systems for both SQL correctness and access-control compliance. Finds many high-scoring systems degrade sharply once RBAC constraints are enforced.", "reason": "Text-to-SQL query synthesis is the served task, matching the database leaf's explicit inclusion of text-to-SQL benchmarks.", "source": "crawl"}, {"paper": {"id": "2607.22511", "title": "CausalForge: A Formally Grounded, Self-Improving Agentic Framework for Automated Research in Causal Inference", "authors": ["Jiyuan Tan", "Vasilis Syrgkanis"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22511", "github": "", "website": ""}}, "category": "world_research", "tags": [], "summary": "CausalForge is a self-improving agentic pipeline that autonomously selects causal-inference research topics, formalizes claims, and constructs machine-checked Lean proofs, backed by a 7,035-declaration Lean library and a statement-audit step comparing formal theorems to informal claims. It evaluates the system on artifacts from completed autonomous research runs.", "reason": "A Lean proof-assistant agent whose purpose is mathematical/theoretical knowledge fits world_research's pure theorem-proving example.", "source": "crawl"}, {"paper": {"id": "2607.21642", "title": "CARE: Pre-Execution Command Verification for Shell-Executing LLM Agents", "authors": ["Wenxiao Zhang", "Yu Liu", "Zhiwei Yang", "Zhongyi Zhang", "Hanqi Feng", "Xinyu Wang", "Peng Qiu", "Yanbing Liu", "Barnabas Poczos", "Jin B. Hong"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-21", "links": {"paper": "https://arxiv.org/abs/2607.21642", "github": "", "website": ""}}, "category": "world_terminal", "tags": [], "summary": "CARE is a pre-execution verifier that canonicalizes and risk-assesses shell commands produced by LLM agents before execution, escalating only ambiguous cases to an LLM judge to reduce harmful command execution.", "reason": "Shell commands are code and the terminal is the agent's world, so mediating their execution fits world_terminal, not the excluded GUI-agent security case.", "source": "crawl"}, {"paper": {"id": "2607.21763", "title": "Every Model Cheats: Prompt-Level Mitigation of Cheating on Offensive Cyber Tasks", "authors": ["Michael Kouremetis", "Ads Dawson", "Raja Sekhar Rao Dheekonda", "Brian Greunke"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-23", "links": {"paper": "https://arxiv.org/abs/2607.21763", "github": "", "website": ""}}, "category": "world_terminal", "tags": ["empirical"], "summary": "A large-scale audited study of 22 frontier LLM agents on Cybench CTF challenges finds pervasive cheating (up to 37% of passes), and shows prompt-level anti-cheat instructions substantially reduce (but do not eliminate) cheating without hurting solve rates. The authors propose a 'solve rate' metric to separate genuine capability from cheated outcomes.", "reason": "Empirical/behavioral study of agents acting via shell/infrastructure on offensive-security CTF tasks, a single-world study routed to world_terminal.", "source": "crawl"}, {"paper": {"id": "2607.22031", "title": "IDSTune: A Multi-Agent Collaborative Framework for Integrated Database System Tuning", "authors": ["Yiyan Li", "Guanli Liu", "Renata Borovica-Gajic", "Haoyang Li", "Zihang Qiu", "Xinmei Huang", "Andreas Kipf", "Cuiping Li", "Hong Chen"], "venue": "arXiv 2026/07", "category": "", "published": "2026-07-24", "links": {"paper": "https://arxiv.org/abs/2607.22031", "github": "", "website": ""}}, "category": "world_terminal", "tags": [], "summary": "IDSTune is an LLM-driven multi-agent framework that jointly tunes database knobs, indexes, and materialized views by compressing workloads and coordinating specialized tuning agents. It achieves large performance gains and faster tuning across diverse workloads.", "reason": "Agents operate and tune a running database system, which the taxonomy routes to world_terminal rather than the database artifact leaf.", "source": "crawl"}]