The goal of this thesis is to study the planning method of a skill based robotic system, and to learn and to present the advantages and the applicability of this method. It’s important for an agent to be able to adapt to the environmental conditions. Also, we would like to use already designed sensor and actuator drivers, in different situations without the necessary configuring, or reconfiguring of the whole systems. For the thesis, I selected a problem based in a real-life application, to simulate and to test the agent’s problem solving capability. The application is simulated on the Jason platform, and we can test the agent’s program in this simulated environment as well. The problem is divided into four scenarios. The main issue in planning of what the scenarios should be was how realistic they were, and to provide a challenge for the agent (reconfiguring). The agent’s task has been divided into subtasks, and from the subtasks skill tree was built. The skill tree includes the subtasks for all four scenarios. The nodes of the tree can be “AND” or “XOR” nodes expressing the decomposability of the robotic skills. The tree can be pruned by the “XOR” nodes. After pruning this will leave a sub-tree based on the skills required by only one of the four scenarios. For a successful pruning, we need the describing parameters of the selected scenario. After the pruning, the leaves left in the tree select the subtasks the agent needs to use in the scenario. An algorithm integrates the selected skill code into the main code of the agent controller. The main code of the robot controller was designed by hand for this task alone. I also designed an algorithm to select the skills required by the parameters of the selected scenario from the .json based skill tree descriptor. The selected skills are formulated as .asl code plug-ins, so the Jason platform can run them. After the .asl is ready, the full program is tested in the actual scenario’s simulation, to see how the agent can perform the task.