An open agent-security benchmark, including the attacks we fail to catch

**Summary:** This open-source project is an agent-security benchmark that helps developers evaluate AI capabilities without starting from a blank model stack, i