This paper proposes a design method for agent interaction protocols (AIPs) in multi-agent systems. The method starts a design process with informal and ambiguous requirements in English, and according to the clearly defined development stages the method gradually converts them into clear specifications expressed in augmented UML. Similar to the object-oriented development, the early stages include extractions of agents, roles of agents, actions, goals, and several types of logical constraints. Based on these information, the successive stages generate skeletons of scenarios, which specify the necessary conditions to achieve a goal. Completing the skeleton with added information from the designer, the skeletons become the final AIP designs. We also show a reuse of design knowledge is possible in each step of the method.