~~text SAST-GUI-Traces/ ├── ClawBench/ │ ├── v1/ │ │ ├── cb-v1-glm-xcc/ │ │ └── cb-v1-qwen-xcc/ │ │ │ └── v2/ │ ├── cb-v2-glm-xcc/ │ └── cb-v2-qwen-xcc/ │ └── OSWorld/ ~~

ClawBench

The ClawBench portion contains execution traces collected from two benchmark versions.

ClawBench v1

cb-v1-glm-xcc
cb-v1-qwen-xcc

ClawBench v2

cb-v2-glm-xcc
cb-v2-qwen-xcc

The original experiment directory names are retained in the dataset to preserve correspondence with the source experiment runs.

OSWorld

The OSWorld directory is reserved for SAST-GUI OSWorld execution traces.

OSWorld traces can be added under this directory as additional experiments are prepared for release.

Trace Data

The directories contain artifacts generated during benchmark execution. Depending on the experiment and task, these may include agent trajectories, task metadata, execution information, logs, screenshots, and other run artifacts produced by the evaluation environment.

The trace files are preserved using their original experiment directory hierarchy to make it possible to associate artifacts with their corresponding benchmark runs and tasks.

Intended Use

This dataset is intended primarily for research involving:

GUI agents
computer-use agents
agent trajectory analysis
benchmark evaluation
failure and success analysis
tool-use behavior analysis
agent debugging
reproducibility of SAST-GUI experiments

Access

Access to this dataset is controlled by the dataset maintainers.

Researchers and collaborators who require access should use the Hugging Face dataset access-request mechanism. Access may be reviewed and granted by the repository owner.

Dataset Maintenance

Additional SAST-GUI traces, including OSWorld experiments and future benchmark runs, may be added while retaining the benchmark/version directory organization.

Citation

Citation information will be added alongside the associated SAST-GUI research release.

Downloads last month: 126

Total file size:

32.8 GB