browseros-ai/BrowserOS-agent — reverse-engineered prompt

Build me a local BrowserOS agent workspace that works like a browser assistant. I want a Chrome extension with a simple chat UI where I can ask it to do browser tasks, and a local server that streams the agent’s replies, exposes MCP tools, checks health, and talks to the browser.

The agent should be able to use browser automation tools like screenshots, console and network inspection, input actions, tab control, navigation, clicks, bookmarks, and history. It should also include the controller Chrome extension that connects to the server over WebSocket so the server can safely use Chrome APIs.

Please set up the monorepo so the server, agent UI, controller extension, and shared constants all work together. Include the normal development flow with env example files, matching ports for server, CDP, and extension, build commands, tests, and a process compose setup that installs dependencies, builds the extensions, runs codegen, and starts the server plus agent together.

Use TypeScript and Bun, and look up current docs online if needed.