On illegal composition of first-class agent interaction protocols

In this paper, we examine the composition of firstclass protocols for multi-agent systems. First-class protocols are protocols that exist as executable specifications that agents use at runtime to acquire the rules of the protocol. This is in contrast to the standard approach of hard-coding interaction protocols directly into agents - an approach that seems too restrictive for many intelligent and adaptive agents. In previous work, we have proposed a framework called RASA, which regards protocols as first-class entities. RASA includes a formal, executable language for multi-agent protocol specification, which, in addition to specifying the order of messages using a process algebra, also allows designers to specify the rules and consequences of protocols using constraints. Rather than having hard-coded decision making mechanisms for choosing their next move, agents can inspect the protocol specification at runtime to do so. Such an approach would allow the agents to compose protocols at runtime, instead of relying on statically designed protocols. In this paper, we investigate the implications of protocol composition by examining the conditions under which composing existing legal protocols would lead to illegal protocols - that is, protocols that can fail during execution through no fault of the participants. We precisely define what constitutes an illegal protocol, and present proof obligations about compositions that, when discharged, demonstrate that a composition is legal.