runablehq/mini-browser — reverse-engineered prompt

Build me a small TypeScript command line browser tool that helps AI agents use Chrome like a simple set of shell commands.

I want to be able to start or restart a debug Chrome window, open a URL, go back and forward, print the current URL, get visible page text, take screenshots, see clickable form controls with coordinates, click, type, fill fields by label or placeholder, press keyboard shortcuts, scroll, hover, drag, and wait for pages to load.

It should also support multiple tabs, simple JavaScript evaluation on the page, console log streaming, page recording to webm, mp4, or gif, and a basic design audit for colors, fonts, contrast, accessibility, and SEO info.

Make it feel like normal Unix tools, commands read arguments, write useful output to stdout, support JSON output where it makes sense, and compose nicely with pipes. Include clear install and local development instructions, plus a short README with examples an agent could follow.