Jev Skill Family Becomes First to Score Perfect 10 on Security Across All Members

A family of five AI agent skills called Jev has become the first group to achieve a perfect 10/10 security score across all members in a skill evaluation directory that rates tools on six dimensions. The five skills — covering classification, evaluation, evidence extraction, workflow design, and action selection — are built to work as a composed system rather than standalone prompt wrappers. Each skill handles one specific type of judgment and returns a decision to the host agent, deliberately avoiding actions like browsing, executing code, or sending data. Notable features include a smoke-test pattern for bulk classification, citation-preserving evidence extraction that reports when no match is found, and an action-selection skill that explicitly separates selection from execution. The open-source repository, licensed under MIT and CI-tested, documents 108 scenarios and 14 recorded input/output pairs for transparency and verification.