In this paper, we propose a method for total knee arthroplasty (TKA) bone pose identification for robot-assisted orthopedic surgery. The TKA procedure presents a unique challenge because of a finite number of cutting trajectories, their relative locations that are not completely known, anatomical constraints and bone placement errors. Our method addresses these challenges by constructing cutting task-specific capability maps which represent the robot's ability to execute the task for various task poses. To identify the femur and tibia bone poses from the map, we develop a feasibility measure which scores the bone poses based on the volume of feasible workspace locations. Our method is successfully tested on the TCAT® surgical robot and the results reveal that it is able to properly identify anatomically feasible bone poses that have enough margin for bone placement errors during the surgery.