Skip to main content
The vocabulary: a target (your agent, registered with a connection config), an assessment run (adaptive attacks against it), findings (validated failures with severity and a lifecycle), and the judge that grades every exchange fail-closed.