alibaba/page-agent ? reverse-engineered prompt

I want to create an AI agent that can be easily added to any webpage. It should act like a copilot living directly on the page, allowing users to control the site with simple text commands. For example, a user could type "fill out the form with my shipping details" or "find and click the login button" instead of manually searching and clicking.

The whole thing should run in the browser using JavaScript, without needing a complex backend or making users install a browser extension. It should just use text to understand the page structure, so we don't need fancy multi-modal models. I'll want to plug in my own API key to use a specific large language model.

Also, for more advanced scenarios, it would be cool to have an optional Chrome extension that lets the agent work across multiple tabs.