Production systems must be more resilient and adaptive due to mass customization and increasingly external disturbances, such as supply chain disruptions or changing policies. As the last chain in the production value stream, assembly systems are especially prone to fluctuations, leading to alternative and more flexible assembly system designs. Online scheduling is a crucial component for dynamically controlling a flexible assembly system. This work presents a modular software architecture that interfaces between online scheduling agents and control systems. A standardized data model of the assembly system allows for exchanging different scheduling agents during the planning or operation phase. Applications are benchmarking competing algorithms, validating scheduling results by comparison, and seamlessly substituting or updating scheduling algorithms. The standardized data model and interface on the assembly system side facilitate the transition between planning and operation. A simulation model can be interchanged with a control system without extra effort to integrate the control system's scheduling agents. Additionally, the modular architecture enables production-parallel simulation to optimize the running system by evaluating and executing alternative scenarios. The long-term assembly system performance can profit from the modular architecture by updating the agent during production with advances in online scheduling algorithms (e.g., machine learning). Furthermore, the modular architecture enables the required resilience and adaptability by fast switching from simulation to real control systems and supporting system optimizations during operation.