Great question, so right now you can do both. It does work better if you simply enter in the url for your task.

For the url generation we do we have safety checks for the urls however it's simply in the prompting. I would love to hear what sort of safety suggestions you have and/or concerns about this sort of experience. Right now we're still figuring out how best to enable people to utilize agents safely.