Model of motor skill pattern is a fundamental component to support feedback mechanism in haptic-assisted motor skill training. Because of the diversity of manipulation tasks in real world, it is a challenge to construct a generic model for various motor skill patterns. Considering fine motor skill mediated by a rigid tool, criteria for a general model of motor skill are identified: generality, quantifiable representation and the capability to support real-time error computation. A hierarchical motor skill model is proposed in a two-level architecture: the spatial-temporal coupling mechanism in task level and the quantified representation in action level. The purpose of task level is to decompose a motor skill into basic elements that can be described with quantified representation. The purpose of the action level is to give quantified representation in each single dimension in the Cartesian Space. Based on the combinations of independent variants, mapping functions in the action level are classified as five basic types. The proposed model can provide a systematical view to compare various motor skills in literatures, and thus can help us to identify unexplored topics in haptic-enabled motor skill training area. Furthermore, examples of two motor skill tasks show that the proposed model can provide a systematic way to design training system for complex motor skills.