we propose a unified framework for integrating planning, scheduling and intelligent prompting into a decision-theoretic model of multi-task assistant for people with cognitive impairments. The principle difficulty inherent in developing a task assistant system is the size of state space that can be handled by the model. We address this issue by modeling the problem as a hierarchical Markov decision process with abstracted actions, described as options. We extend the expressive power of the model through a wide repertoire of heterogeneous options to cope with highly dynamic and partially observable environment. Another challenge in developing assistive technology is to design effective interaction strategies adaptive to individual user’s needs and preferences. We will investigate different aspects regarding the implementation of interaction behavior, including regular prompt and inquiries (either for choice making or resolving state uncertainty), and explore the methods of interaction strategy learning.