The browser subagent can’t scroll. Not on every page, but on enough pages that it’s a real blocker for any workflow that needs to see content below the fold.
Tested this on a page with a long list of items extending past the viewport.
Steps to reproduce:
- Use
browser_subagentto navigate to any page with scrollable content (any page with content below the fold) - Ask the agent to scroll down using the scroll tool
- Watch the page. It either doesn’t move or it moves briefly and snaps back to the top
The agent reports the scroll happened. The page didn’t actually move. window.scrollY sometimes updates but the rendered viewport stays pinned at 0.
Tried window.scrollTo(0, 1500) via JavaScript execution too. Same result. The internal scroll position changes but the visible page bounces right back.
I can scroll the same page manually with zero issues. It’s only the agent’s scroll commands that fail. This isn’t site-specific scroll hijacking because manual scrolling works fine on the exact same page in the exact same browser instance.
The workaround right now is to extract everything from the DOM via JavaScript without ever scrolling. That works for data, but any task that requires the agent to visually locate or interact with elements below the fold is stuck.