README.md

metadata

pretty_name: GUI-vs-CLI A Unified Benchmark
task_categories:
  - text-generation
language:
  - en
size_categories:
  - n<1K

GUI-vs-CLI: A Unified Benchmark

This dataset contains task descriptions and verification specifications for 440 desktop software tasks from the GUI-vs-CLI benchmark.

The Hugging Face dataset is intended for browsing and lightweight programmatic access to task descriptions. Full runnable task assets, environment files, and execution code are maintained in the GitHub repository.

Files

• data/tasks.jsonl: one task per line.

Fields

Each row contains:

• id: unique task identifier.
• app: target desktop application.
• task: natural-language task instruction.
• env: referenced runtime files and sandbox paths, when applicable.
• verification: verification commands and expected checks.
• github_task_path: path to the corresponding task directory in the GitHub repository.

Task Count

Total tasks: 440

Applications

• audacity: 24
• chrome: 17
• cloudcompare: 23
• drawio: 15
• freecad: 26
• gimp: 19
• godot4: 19
• krita: 17
• libreoffice_calc: 36
• libreoffice_impress: 32
• libreoffice_writer: 39
• musescore3: 25
• obs: 18
• obsidian: 23
• renderdoc: 41
• shotcut: 20
• zoom: 20
• zotero: 26

Usage

from datasets import load_dataset

dataset = load_dataset("rebeccazzzz/GUI-vs-CLI-A-Unified-Benchmark", split="train")
print(dataset[0]["id"])
print(dataset[0]["task"])

Notes

This Hugging Face repository intentionally excludes binary environment assets such as spreadsheets, project files, images, and databases. Use the github_task_path field to locate the full runnable task package in the GitHub repository.