This message was deleted.

I am facing 2 main issues with seeact. 1. elements like iframes are not being captured. For example here, this pop-up window is an iframe component, and this isnt part of the HTML dom tree. So the agent isnt able to to figure out to even close the window before proceeded. 2. SeeAct passes the entire HTML dom tree as an input, not just the visually detectable areas of the screen. Screenshot captures the full page and not the visual section. Because of this the agent ends up performing some random actions.