xlang-ai/OSWorld ? reverse-engineered prompt
Reverse engineered prompt
Build me a tool that lets me run OSWorld style desktop tasks inside a virtual computer and test how well an AI agent can use a real GUI to finish them.
I want a simple command line workflow where I can start an environment, load a task, run an agent, and then see the result and score at the end. It should support the common ways people run these desktop environments, including local virtual machines and container based setups, and it should be able to save logs, screenshots, and run results so I can compare different agents later.
Make it easy to try with a quick start script, and include a clean way to run one task or a full batch of evaluations. If you need to check current setup details or platform docs online, go ahead and do that. The main goal is a working benchmark runner for computer use agents, not a fancy app.
Are you gonna build this?
make sure you review the code using coderabbit