A proper browser automation CLI that talks to Chrome over CDP and treats each action as a small Unix tool. You get the usual suspects like navigation, screenshots, and text extraction, but also genuinely useful stuff like snap (lists clickable elements with coordinates), fill (smart form filling by label or placeholder), and even design audits. The observe-then-act loop is clean: snap the page, grab coordinates, click or type at those coordinates, wait for the dust to settle, repeat. Works well for agents because everything is coordinate-based rather than brittle selector hunting. One recording at a time, viewport is fixed at 1024x768, and you'll need to scroll and snap again to find elements below the fold. Solid choice if you need an agent to actually drive a browser without fighting CSS selectors.
npx skills add https://github.com/runablehq/mini-browser --skill mini-browser