This paper proposes a novel trust-based impedance control scheme based on task performance metrics and faults that allow a robot to act as a supervisor for its human partner in a cooperative human-robot task. A dynamic Trust model is used to modulate the robot stiffness as a function of the user performance. The task metrics are accuracy, forces applied by the user and the time taken for completion of the task. Results show that the proposed control scheme results in lower forces applied by the user while simultaneously ensuring accuracy of the task. The proposed methodology can be expanded to train a novice user to match the performance of a professional user.