@agentic.james: Start thinking of agent skills as little pieces of agentic software, because that is what they are and they might be the future of computer work. LLMs are natural language computers, and skills are the natural language software that runs on them, the same way traditional software runs on traditional computers. Instead of writing your logic in if statements and while loops you write it in plain language for the model to interpret, and there is almost no limit to the logic you can pack into one skill file. You can even drop real scripts inside a skill for the agent to execute with its bash tools. Granular skills plug into bigger skills exactly like software modules. Once you see skills as agentic software the whole thing clicks.