How does the user know what the page is capable of in order to be able to instruct the agent?