Skills were always prototypes
The models are absorbing what we used to write down, which is exactly what a good prototype is for.
Not long ago, getting good results from a model meant building around it. You wrote skills to tell it how to approach a task, set up a harness to feed it the right context and tools, and kept a small library of instructions for all the things it wouldn't do on its own. Ask before building. Check the work. Format the output properly.
A lot of that is quietly becoming unnecessary. The models ask clarifying questions without being told to, the products around them produce proper documents and decks out of the box, and context now carries between sessions without anyone maintaining markdown files by hand. Things people carefully spelled out in a SKILL.md a year ago are now just what happens by default, and a good part of the harness comes pre-built.
Insight: A skill is a written-down workaround. It is worth roughly as much as the gap it fills, and the gaps are closing.
It's tempting to read that as wasted effort, but I think it's the opposite. The skills and harnesses people built were prototypes of what the platforms were missing, and the platforms caught up because enough people showed them where the gaps were. That is exactly what a prototype is supposed to do. Nobody in product work is upset when the prototype doesn't become the final product.
The more lasting value was never the file anyway. Writing a good skill forces you to be precise about how a piece of work should be done: which questions matter, what order the steps come in, what the output looks like when it's right. That understanding stays when the file becomes redundant, and it shows up in how you brief, how you prompt and how quickly you notice when something is off.
What's left worth writing down is the personal part. Preferences, constraints and ways of working that only make sense in your context. The models can't learn those unless someone tells them, because there's nothing else to learn them from. So skills aren't going away, they're just getting smaller and more specific, moving from "here's how to do this" to "here's how I like it done".
I still think they're worth building, mostly for what they teach you about the work, even if you end up not needing them for very long.