This paper introduces a computer-assisted tele-operation system to improve task performance for remote manipulation tasks. The system shares control between a human operator and computer assistance. It uses two units: an action recognition unit that can evaluate haptic data at high sampling rates and handle changes from assistance, and an assistance unit that provides context-specific help. The system is evaluated on a hard drive repair task, with the action recognition achieving 65% accuracy. A user study found the assisted system significantly reduced human effort and risk of damage compared to an unassisted one.