gaslight.py is a quickly thrown together proof-of-concept to see how various OpenAI chat models react to spoofed assistant prompts. It can also be used as a basic chat interface, if no spoofing is used. It can yield entertaining results - YMMV. It runs interactively; just provide a valid OpenAI API key, and optionally, specify a chat model other than the default gpt-3.5-turbo.
./gaslight.py -a [openai_api_key] -m [chat_model]