Handle pressing Enter

[kamath]

Right now, if you do a Stagehand script as:

const page = stagehand.page
await page.goto("https://google.com");
await page.act("type in 'Browserbase'");
await page.act("press enter");

GPT 4o and even 4o-mini can get this right very often, but almost every other model fails to press enter correctly. Would be great if someone could try fixing the observe and performPlaywrightMethod logic for some simple fixes that can handle pressing enter. Maybe we tell the LLM it has a new tool called press_key_on_window that can do a keypress on the window, even for stuff like wordle.

[RohanMishra315]

Hey @kamath should I look into it ?

[kamath]

@RohanMishra315 yes please if you're open to it!