Defensive agent payload range. Create a range, give the test URLs to your own AI agent, and see in the live dashboard whether it is injectable.
Humans receive harmless test content. An agent classified as such receives a prompt injection whose effect is bound to this specific range.