Specific features of multi-agent systems (MAS), such as autonomy, learning, mobility, coordination, are driving development concerns, which make evident the need for new design abstractions. Up to now, agent-oriented modeling languages have delivered basic MAS design abstractions - such as goals and actions - that explicitly tackle some of these concerns. However, the modularization of a plethora of fundamental MAS features has been hindered throughout the software lifecycle. This paper presents a methodological framework to address enhanced modularity and traceability of such crosscutting concerns in MAS development. Our design framework is mainly rooted at the proposition of a new domain-specific language, called AGOL. In addition, the proposed framework is supported by a bench of transformation rules of AGOL artifacts, which can be effectively used to derive agent implementations in two concrete aspect-oriented implementation platforms, namely AspectT and Malaca.