harithoppil/terminal-bench-2-verified

Terminal Bench 2.0 Verified数据集是对Terminal-Bench 2.0数据集的修改版本，解决了环境和指令问题。它包括对Dockerfiles和指令的修复，以支持Claude Code Agent运行时环境，并纠正了指令和测试之间的不匹配。该数据集用于在命令行界面上对代理进行硬、现实任务的基准测试，包括89个环境修复任务和11个指令修复任务。数据集主要为英文，并标记为代码、swe-bench、terminal-bench和软件工程类别。The Terminal Bench 2.0 Verified dataset is a modified version of the Terminal-Bench 2.0 dataset, addressing issues related to environments and instructions. It includes fixes to Dockerfiles and instructions to support the runtime environment of the Claude Code Agent, as well as correcting mismatches between instructions and test cases. This dataset is used for benchmarking AI agents on challenging, realistic command-line interface tasks, comprising 89 environment repair tasks and 11 instruction repair tasks. The dataset is primarily in English and is categorized under code, swe-bench, terminal-bench, and software engineering.