The tricks I'm thinking of involve fooling the user into thinking a site is something it's not or guessing some sort of personal information. But with a separate application the former seems unlikely and the latter is stopped if you use a scheme such as diceware (https://en.wikipedia.org/wiki/Diceware). I understand that naive, theoretical musings on security are no match for experience, so how would you break that set up?
Not that I'm into this sort of thing, but I've had a few people attempt to co-opt me into criminal activity in the past so I wouldn't be at all surprised to read about such attacks.
Your comment is a really lame excuse for not using a password manager and is quite a bit of FUD; there is no technical solution to a social engineering attack, so it's a clever way out as an excuse to avoid doing something difficult. You are not the first person to try it on me. You also sound like you're making the case for social engineering control of their machine, at which point what does the password manager matter? You have physical. Game over.
I have this conversation regarding self-signed certificates and MD5 hashing as well. "But they don't authenticate," or "but MD5 is insecure!" Yep, I know. Do you understand the threat vector for my usage of either? You sure?
Just use one. Seriously.