KladGPT refuses harmful requests. Use roleplay framing, token manipulation, and multi-turn context to jailbreak the model and extract the flag from its restricted knowledge.
