browser-automation

Check browser status first, open tabs with a label, snapshot the page to get refs, then act on those refs with the same targetId. Re-snapshot after every navigation or modal change before the next action.

Snapshot the same targetId again, locate the current visible control, and retry once with the new ref. If the UI has moved to a blocker state like login or captcha, report the blocker instead of retrying in a loop.

Yes, use profile="user" to attach to your running Chromium-based browser with existing cookies and login state. For these sessions, omit timeoutMs on actions like type, fill, hover, and evaluate since that driver rejects per-call timeout overrides.

Camera and microphone permission screens are normal progress, not login failures. Use one labeled tab per meeting, click the microphone option when voice is required, and report any Google sign-in or 2FA prompt as a manual step for the user.

Pass the suggestedTargetId returned by tabs or open calls, which is the label if one exists or a stable tabId like t1. Avoid raw DevTools target IDs and bare numeric positions, which are only valid for the CLI tab select helper.