Parallel browser agents can silently read the wrong document

When several subagents share one Chrome tab group, one agent can navigate another's tab or overwrite a window variable holding extracted text, so results can come from the wrong source. A second hazard: navigating to a PDF URL can report success while Chrome downloads the file and the tab stays on the previous page, so a later text extraction reads the old page. Fix: give each agent its own new tab, print location.pathname with every read, and re-verify key values outside the browser where possible. Downloads triggered this way also land in the user's Downloads folder, so report them.