With introduction of new technologies in the operating room like the da Vinci Surgical System, training surgeons to use them effectively and efficiently is crucial in the delivery of better patient care. Coaching by an expert surgeon is effective in teaching relevant technical skills, but current methods to deliver effective coaching are limited and not scalable. We present a virtual reality simulation-based framework for automated virtual coaching in surgical education. We implement our framework within the da Vinci Skills Simulator. We provide three coaching modes ranging from a hands-on teacher (continuous guidance) to a handsoff guide (assistance upon request). We present six teaching cues targeted at critical learning elements of a needle passing task, which are shown to the user based on the coaching mode. These cues are graphical overlays which guide the user, inform them about sub-par performance, and show relevant video demonstrations. We evaluated our framework in a pilot randomized controlled trial with 16 subjects in each arm. In a post-study questionnaire, participants reported high comprehension of feedback, and perceived improvement in performance. After three practice repetitions of the task, the control arm (independent learning) showed better motion efficiency whereas the experimental arm (received real-time coaching) had better performance of learning elements (as per the ACS Resident Skills Curriculum). We observed statistically higher improvement in the experimental group based on one of the metrics (related to needle grasp orientation). In conclusion, we developed an automated coach that provides real-time cues for surgical training and demonstrated its feasibility.