Rainbow Team is a multi-role adversarial evaluation framework designed to probeAI systems across robustness, safety, reasoning, continuity, and multimodalconsistency. It defines a structured protocol consisting of color-codedadversarial roles, layered stress intensities, fault-injection operators,counterfactual probes, and multimodal inconsistencies, enabling systematicpressure-testing of models under diverse adversarial conditions. The framework evaluates long-range reasoning stability, safety-gate integrity,continuity drift, counterfactual sensitivity, multimodal grounding, and failurefingerprints. Rainbow Team introduces standardized evaluation cycles, protocolizedtesting phases, redundancy-free adversarial roles, and reproducible evaluationmetrics. It supports LLMs, multi-agent systems, multimodal models, agents,tool-using models, and cognitive operating systems such as CodexOne. This release provides the full Rainbow Team specification, methodology,layered testing protocol, adversarial role definitions, evaluation metrics,and fault-injection operators. It forms the basis for future systematic,governable, and interpretable evaluation of advanced AI systems.