georvn7/super-debug-v2

super-debug-v2是一个合成的软件调试轨迹数据集，由名为hen的AI编码代理生成，专注于C/C++项目。这是super-debug的第二版，相比第一版，此版本包含了通过完整测试套件的三个调试运行轨迹。数据集包含三种主要训练文件：默认的无助手思考字段的SFT文件、保留助手思考字段的SFT文件以及DPO偏好数据文件。数据集旨在用于训练或评估基于实际调试行为的编码代理，研究合成轨迹蒸馏和障碍识别，以及调试器中心的工作流研究。数据集的局限性在于它是合成数据，非人工标注，且反映了hen的结构和动作词汇，主要面向C/C++项目。

super-debug-v2 is a synthetic dataset of grounded software-debugging trajectories generated with hen, a stateful long-horizon AI coding agent for C/C++ projects. This is the second version of super-debug, which includes trajectories from three full-suite debugging runs that pass the complete test suite. The dataset consists of three main training files: a default chat-style SFT file without assistant thinking fields, a matching SFT file with thinking fields preserved, and a DPO preference data file. It is intended for training or evaluating coding agents on grounded debugging behavior, research on synthetic trajectory distillation and blocker identification, and studying debugger-centered agent workflows. Limitations include that it is synthetic data, reflects the structure and action vocabulary of hen, and is currently C/C++ oriented.