A common characteristic of modern Deep Learning approaches is that they involve training models “from scratch” to solve specific tasks. This approach, while effective, poses significant challenges especially in the field of Reinforcement Learning (RL) leading to unsustainable computational efforts. In fact, RL agents must learn optimal policies without any prior knowledge, leveraging only their experience with the world. Furthermore, lifelong learning agents continually adapt to different scenarios and objectives over time. In most cases, these challenges are dealt with without exploiting previously learned skills or policies to a meaningful extent. The following poster aims at presenting a distinctive perspective on how to address these challenges by redefining the concept of skills within the RL framework.