In order to describe how a structure - System Evolution Directory (SED), can be used to facilitate the evolution of multi-agent systems, we first introduce a case study on citation finding. We use this case study to explain the components within a SED and the process to build a SED. A unified and simple model directory is used to describe the life cycle of a system and a goal model to describe the meaning among system artefacts. Developers can not only get all necessary information ranging from high level system requirements to low level source code from the model, but also can understand semantic connections between system artefacts. Different versions of a system can be recorded easily in the structure. We believe the structure has potential for further speeding up and automating the process of system evolution. All in all, the structure complements current agent-oriented methodologies.