Compass TaskWindow

Lets your Compass coding agent open, read and test web pages in its own tabs, in a browser you watch.

Compass TaskWindow gives your Compass coding agent a browser to work in while you watch. Ask Compass to check a page. It opens the page in its own tab group, reads it as structured text, clicks, types and fills forms like a user, takes screenshots, and reads the console and network log. Then it tells you what it found. What it can do - Open pages in one labeled tab group per Compass session: "Compass task 1", "Compass task 2" - Read elements, text and tables, and act on them by reference, CSS selector or screen position - Capture a page or one element, and record steps as an animated GIF - Show each page's console and network log, downloads included - Wait for text or an element to appear or go away, and close its tab groups when done You stay in control - It reads and controls only the tabs it opened. Your own tabs are never touched. - Every page it drives is framed in its session's color, with a Stop button on the page. - The toolbar button stops the session in the tab in front of you. How it connects TaskWindow talks only to a small helper program on your computer, which Compass installs with /taskwindow setup, through Chrome native messaging. It has no server of its own, needs no account and collects no analytics. Page content goes to the Compass agent you run, and from there to the AI provider you configured. Requires Compass with the Compass Superpowers bundle, on macOS or Linux.