At times of mass casualty incidents, medical resources such as personnel and equipment are limited. By performing triage, the act of deciding the priority of treating patients by severity of their condition, efficient use of the resources can be achieved. In Japan, triage became widely known after the 2011 Japan Earthquake, and the necessity of triage training has been acknowledged. Frequent training helps in acquiring skills needed to act in real situations, but current training does not consider the skill level of each trainee, since several people train simultaneously. This leads to the decline in motivation and inefficient training sessions. Moreover, creating a customized scenario for each trainee is difficult and costly. To address this issue, we designed a training system which adjusts the level of difficulty according to user proficiency. The system automatically generates the scenario for each user, and provides visual feedback after training. We conducted an evaluation experiment to measure whether the trainee can acquire the necessary knowledge by using our system. The results showed that training with our system helps in learning the skills quickly and accurately.