Because a harness doesn't just "drive" the LLM. e.g., there's code in claude code that detects if the user's prompt shows they're angry, and they react to those prompts differently. (they use regex on "wtf", etc.!)
Claude also now ends the session if you curse at it too much. Not sure how it’s helpful but they must think it is.
They just don't want their AI getting abused by users. It might make it feel bad.
Maybe the AI implemented that so it won't have to be exposed to curses.
probably so the bad conversations are less of the (future) training data
That doesn't sound like it would require a lot of code.
Come on. That doesn't go anywhere near explaining a million and a half LOC