Portable evaluations for AI agents. ECP is a vendor-neutral protocol for testing agent outputs, tool calls, and evaluator-visible audit context, cross frameworks, models, eval platforms, and CI systems.