The Control Script doc says:

My Suzette bot had a control script that would analyze your interactions
with her. If you were friendly, agreeable, interested in the conversation, she got friendlier.
Until she reached a point where she went neurotic and felt she wasn’t deserving of your
attention. This became clear in her conversations. Likewise if you were hostile,
disagreeable, and bored with her, she drifted into paranoid. She wondered who might be
listening in on the conversation, why you asked the questions you did. She even began to
speculate how she might kill you. It’s all in the control script she had (and topics she had
that got invoked to simulate paranoia or neurotic affection).

I would like more information on how this works.
Is there a source of Suzette or smiliar bots I could look at?

No. I have not released her source.

Analysis of user input detected relevant data. Long inputs express interest in a topic. Short ones generally dont (running average). User insulting you obviously indicates hostility as does saying no a lot.  User agreeeing or complimenting her indicates liking. User can also explictly say they are bored or want to change topic. But doing that a lot is disrespectful. Whereas a user completing a topic shows a liking for what the bot is saying and an interest.

Permute a bunch of measures to decide how to change the personality. When it crosses a threshhold (with hysteris) then it turns on a variable corresponding to the persona, which randomly adds in front of whatever else the system was going to say, a sentence from the persona.