Writeup for “Exploiting insecure output handling in LLMs”