Demonstrating or pre-programming solutions to all possible tasks that a robot may encounter is unrealistic in scenarios that go beyond the structured industrial environment.Many tasks can be seen as sequences of simple sub-tasks. We propose an approach that allows robots to solve various tasks by sequencing elementary skills. These skills are represented by simple previously learned movement primitives. Each primitive only performs a limited action. Solutions for unseen problems are generated by combining these simple primitives into sequences, here using the A* graph search algorithm.