Browser and web tasks
An Agent equipped with a browser can read pages, navigate, enter information, and inspect results. A lightweight Agent without a desktop is not ready for GUI browser work; confirm its environment first.
You need the appropriate site account and permissions, a ready browser, and access to the target URL. Treat login sessions, cookies, and downloads as sensitive data. Ordinary users cannot select every runtime; explain the required task to an administrator when adjustments are needed.
Request a task
- Give the target URL and expected result.
- Specify whether reading, input, submission, downloads, and external transfers are allowed.
- Start with a small read-only check.
- Ask for source URLs, actual changes, and failed or unexecuted steps.
Follow the browser and CLI guide. Stop and ask the user when login or human verification is needed.
Verify the target Agent's tools and access before requesting browser work. A separate Portal AI Chat menu is not a prerequisite.
Instructions are not technical access controls. Follow organization access and external-transfer policies.