The application of machine-learning solutions to movement assessment from skeleton videos has attracted significant research attention in recent years. This advancement has made rehabilitation at home more accessible, utilizing movement assessment algorithms that can operate on affordable equipment for human pose detection and analysis from 2D or 3D videos. While the primary objective of automatic assessment tasks is to score movements, the automatic generation of feedback highlighting key movement issues has the potential to significantly enhance and accelerate the rehabilitation process. While numerous research works exist in the field of automatic movement assessment, only a handful address feedback generation. In this study, we propose terminology and criteria for the classification, evaluation, and comparison of feedback generation solutions. We discuss the challenges associated with each feedback generation approach and use our proposed criteria to classify existing solutions. To our knowledge, this is the first work that formulates feedback generation in skeletal movement assessment.