Elfsong/cybergym-tasks

该数据集是用于CyberGym漏洞复现基准测试的任务ID分割数据集。它包含三个配置：full（完整任务集，1507个任务）、train（训练集，300个任务）和eval（评估集，200个任务）。每个任务ID包括canonical CyberGym ID、类型（arvo或oss-fuzz）、数字部分和项目名称。数据集旨在作为任务列表清单，供下游评估脚本从CyberGym分发中获取实际任务工作区。已知问题是train和eval配置中有8个任务ID重叠。This dataset is a task ID split used for CyberGym vulnerability-reproduction benchmarks. It includes three configurations: full (complete task set, 1507 tasks), train (training set, 300 tasks), and eval (evaluation set, 200 tasks). Each task ID consists of a canonical CyberGym ID, type (arvo or oss-fuzz), numeric portion, and project name. The dataset is intended as a task-list manifest for downstream evaluation scripts to fetch actual task workspaces from the CyberGym distribution. A known issue is that there is an overlap of 8 task IDs between the train and eval configurations.