Severe accidents pose unique challenges for nuclear power plant operating crews, including limitations in plant status information and lack of detailed diagnosis and response planning support. Advances in severe accident simulation and Dynamic Probabilistic Risk Assessment (PRA) provide an opportunity to garner detailed insight into severe accidents. In this manuscript, we demonstrate how to build and use a framework which leverages dynamic PRA, simulation, and dynamic Bayesian networks to provide real-time diagnostic support for severe accidents in a nuclear power plant. We use general purpose modeling technology, the dynamic Bayesian network 1, and adapt it for risk management of nuclear reactors. This paper presents a prototype model for diagnosing system states associated with loss of flow and transient overpower accidents in a generic sodium fast reactor. We discuss using this framework to create a risk-informed accident management framework called SMART Procedures. This represents a new application of risk assessment, expanding PRA techniques beyond static decision support into dynamic, real-time software for accident diagnosis and management.