From Stakeholder Intentions to Agent Capabilities

Agents behavior at run time is influenced by the environment within which they operate, so designing agents able to behave in an effective way should benefit from a deep knowledge of that environment, including the stakeholders who have specific needs and expectations from the agents being designed. The specification of agents behavior is typically addressed during detail design, a step which precedes implementation. Analysis of the environment, on the other hand, is performed earlier, during domain and system requirements modelling, which rest on environmentand problemoriented abstractions. This motivates the need to support traceability between detailed design models of agents behavior and domain models and stakeholder requirements, and viceversa. Moreover, evaluation of agent behavior (i.e., is the agent doing what was intended?) is typically considered a run-time issues. To what extent this evaluation can be done at design time is still an open issue. Our goal in this paper is to address the problem of better linking requirements analysis to detailed design by focusing on the concept of agent capability and on capability modelling in Agent-Oriented Software Engineering. In particular, we revisit the definition of capability in Tropos, referring to definitions of this concepts proposed in current literature and refine the Troposdevelopment process in order to point out how capability specification can result from the integration of various analysis strategies. An example is used to illustrate.