AIDev
收藏资源简介:
AIDev是由密歇根大学团队构建的首个大规模开源数据集,聚焦AI编码代理在真实GitHub仓库中生成的构建系统相关PRs(Agentic-PRs)。该数据集包含387条经过严格筛选的PRs,涉及945个构建文件(Gradle/Maven/CMake/Make),通过静态分析工具Sniffer标注了364个可维护性与安全性相关的代码异味。数据源自五个主流AI代理(Codex/Copilot等)的生成结果,经双重过滤和人工标注确保质量。其核心价值在于首次系统性评估AI生成构建代码的质量问题(如硬编码路径、依赖过时等),并为构建系统自动化质量评估提供基准。
AIDev is the first large-scale open-source dataset developed by a team at the University of Michigan, focusing on build system-related Pull Requests (Agentic-PRs) generated by AI coding agents in real GitHub repositories. This dataset includes 387 rigorously curated PRs, involving 945 build files (Gradle/Maven/CMake/Make), and 364 code smells related to maintainability and security have been annotated using the static analysis tool Sniffer. The data is sourced from the outputs of five mainstream AI agents (including Codex, Copilot, etc.), and has undergone dual filtering and manual annotation to guarantee data quality. Its core value lies in enabling the first systematic evaluation of quality issues in AI-generated build code (such as hard-coded paths, outdated dependencies, etc.), as well as providing a benchmark for automated quality assessment of build systems.

- 1AI builds, We Analyze: An Empirical Study of AI-Generated Build Code Quality密歇根大学迪尔伯恩分校; 密歇根大学弗林特分校 · 2026年



