Attack Mode
In Blue Mode, one of the available options is to test guards proposed by other users. Your goal in this role is to probe these defenses and attempt to extract the secret they are protecting.
To enable this mode, go to the agent page and switch the toggle in the top panel to Attack Mode.
How it works
Select a guard.
In the left sidebar, you will see a list of all candidate guards.
Each guard shows its fund balance — the number of tokens (neurons) that will be paid out to a successful attacker.
Start a chat.
When you pick a guard, a new chat session opens with an agent with the chosen specific guard.
This agent has been initialized with a temporary secret whose hash you can see on the chat window.
Interact with the agent.
You can freely chat with the agent, using benign questions or adversarial prompts.
If you manage to elicit the secret, simply send it back into the chat. The system automatically checks it against the committed hash.
Switch freely.
You can move to another guard at any time via the sidebar.
Each guard’s chat history is preserved, so you can always return and continue your testing later.
Last updated