Testing how well models handle RP-style prompts (actions + dialogue)
I’ve been working on an AI chat platform where users can write in roleplay format, like:
leans against the wall
“So what now?”
Instead of standard prompt/response structure.
Trying to get models to:
- understand mixed narration + dialogue
- stay in character
- respond in the same format
Curious if anyone here has worked on:
- prompt structures for RP
- handling asterisk actions reliably
- maintaining tone consistency over multiple turns
There’s an early version live if anyone wants to test behaviour:
Would really appreciate insight on where this kind of prompt handling usually breaks.