web-infra-dev/midscene ? reverse-engineered prompt

Build me a tool for end to end testing that lets me control apps with plain English instead of selectors. I want it to work like a visual assistant that looks at screenshots, clicks, types, scrolls, and checks what is on the screen the way a person would.

It should work for web apps first, and be able to support mobile and desktop too if possible. I want simple commands like search for something, filter results, wait until a message appears, and confirm a button or error is visible. Please include a clean way to run these tests in an existing browser test setup, plus a nice HTML report that shows screenshots, actions, and what the agent decided at each step.

If it helps, look up current docs online and make the setup feel realistic and easy to use. I also want a small testing framework style workflow so I can mix normal code with natural language test steps.