The paper describes a system platform for virtual human agent simulations that is able to coherently manage the shared virtual environment. Our “agent common environment” (ACE) provides built-in commands for perception and for acting, while the in-between step of reasoning and behavior computation is defined through an external, extendible, and parameterized collection of behavioral plug-ins. Such plug-ins are of two types: the first type defines agent-object interactivity by using a feature modeling approach, and the second type controls the reasoning and behavior of each agent through scripted modules. The system is analyzed and a simulation example integrating some modules with a Lisp behavioral system is shown