SEC-bench/SEC-bench-Pro

SEC-bench-Pro是一个基准数据集，专注于JavaScript引擎（V8和SpiderMonkey）中的真实世界安全漏洞。数据集包含183个实例，其中V8（Chromium）有103个实例，SpiderMonkey（Firefox）有80个实例，涵盖24种不同的漏洞类型（如类型混淆、释放后使用、沙箱绕过、越界读/写等）。每个实例都包含一个已验证的漏洞，附带Docker可复现环境、详细的技术描述（包括根本原因分析和利用潜力）以及真实修复补丁。数据字段包括实例ID、项目、错误ID、Docker镜像名称、工作目录、验证二进制文件、命令选项、目标源文件、目标子目录、漏洞类型、错误类型、描述、漏洞奖励程序价值和修复补丁文本。该数据集旨在用于安全研究和基准测试，支持通过HuggingFace数据集库加载和过滤。

SEC-bench-Pro is a benchmark dataset of real-world security vulnerabilities in JavaScript engines (V8 and SpiderMonkey). Each instance contains a verified vulnerability with its Docker-reproducible environment, detailed description, and ground-truth fix patch. The dataset includes 183 total instances, with 103 for V8 (Chromium) and 80 for SpiderMonkey (Firefox), covering 24 distinct vulnerability types (e.g., type confusion, use-after-free, sandbox bypass, out-of-bounds read/write). Fields include instance ID, project, bug ID, Docker image name, work directory, verification binary, command options, target source files, target subdirectory, vulnerability type, error type, description, vulnerability reward program value, and fix patches. It is designed for security research and benchmarking, and can be loaded and filtered via the HuggingFace datasets library.