webbrain-one/webbrain ? reverse-engineered prompt

Build me a browser extension for Chrome and Firefox that puts an AI assistant in a side panel next to the current tab. I want to be able to ask questions about the page I’m on, get summaries, find links or info on the page, and also tell it to do tasks like click buttons, type into forms, navigate to pages, upload or download files, and fill things out for me.

It should support three modes, one that only reads, one that can act on the page, and one that also exposes extra page details for debugging and editing. Let the user choose between a local model or a cloud model, and make the default work without any setup if possible. Include a clear permission flow before anything important happens, and make the agent able to plan its actions before running them. If you need to look up current extension or browser docs online, go ahead.