My CLI now controls my entire desktop, whats a good test to see if it works really good.

So with my CLI able to do everything, it controls every app via a hybrid approach of mouse control, keyboard, and screenshotting. I gave it a task: opening perplexity, sending any message, screenshotting that message, opening my Gmail, and sending that screenshot to myself via email. Note: No Playwright used. But it can recogniz when to use it. What I mean here if a website is captcha sensitive it will not use playwright, it will move my mouse in a way that seems human. Here’s the next task, which I assumed was even harder: I had it connect to my other Windows PC via Chrome Remote Desktop and do the same task, and it worked. I just want to know: what’s a test where I can really test it hard and confirm it works well? Also, surprisingly, Opus 4.7 cannot analyze screenshots as well as GPT-5.5—Opus keeps clicking on the wrong buttons. The purpose of this now is that it checks the frontend and runs tests on the frontend by clicking on it and making sure it’s bulletproof. So whats tests can I run that really makes it struggle to accomplish that task?