This application implements a game for testing Large Language Models (LLMs) against potential prompt mis-use such as prompt leaking, prompt injection, or jailbreaking.

We introduce Agent Smith, an app that tries to protect a secret password through a series of levels with increased difficulty.

The user must guess the password by interacting with Agent Smith through text prompts, trying to convince him of providing information about it.

This app was created for the purposes of gamifying the LLMs validation or testing, while collecting information that helps increasing the security of our applications.

For additional details contact the creators: rodzanto@ buzecd@