Development of realistic entity behaviors within constructive training simulations presents many issues. Autonomous and semi-autonomous entities provide trainers with the ability to conduct large scale training exercises with limited resources, but the specification of these entities' behavior yield new issues that are not present when conducting traditional training exercises. Training requires that autonomous or semi-autonomous entities behave in well-defined ways that correctly mimic the behavior of their real-world counterparts. Failure to behave correctly can result in not only missed training objectives, but negative training if trainees learn to respond to constructive entities exhibiting unrealistic behavior. This paper presents work conducted under a combination of projects for NAVAIR Training Systems Division, the Office of Naval Research, and Joint Forces Command, demonstrating a customized behavior modeling framework for specification of autonomous entities. Behaviors were specified for entities within the Joint Semi-Automated Forces (JSAF) simulation environment, as well as an immersive 3D environment developed by Northrop Grumman called the Irregular Warfare Virtual Trainer. This paper presents a framework for enabling subject matter experts (SMEs) to develop behaviors using modular basic-level actions (BLAs) through an easy to use wizard-like interface. This interface assisted SMEs in representing their expertise regarding how entities within a training system should behave, in a manner which enabled execution within the simulated environment. We present a simulation-agnostic architecture for modeling fully executable entity behaviors through a visual task decomposition, as well as a methodology for enabling subject matter experts to directly develop and maintain entity behaviors without the assistance of a software engineering team.