With increasing sophistication in modeling interpersonal interactions, including cultural and social factors, it is now possible and desirable to bring together state-of-the-art human behavior models (HBMs) developed through different approaches in a framework that facilitates interoperability, composability, and comparison. Such a framework is needed in order for game or simulation environments to allow heterogeneous agents (synthetic characters driven by HBMs) to interact with each other and to interact with human players, even though their information needs and operating premises may differ. An important part of such an interoperability framework is an agent message system supporting multiple levels of abstraction. To demonstrate the viability of such a framework, we created an environment that supported synthetic characters interacting with each other and a human player in a common scenario. The agents were provided by different researchers built upon different technical approaches, yet were able to communicate through messages featuring multiple levels of abstraction to describe actions and events. In this paper, we describe our choices of message abstraction levels, what we developed, and the architecture of the framework that we used to mediate the interaction among all the participants, synthetic and human.