Recently, many people on Twitter have been hyping up Skills.

Actually, I approached the experience and learning with skepticism from the start. There are all kinds of claims, saying that it suddenly allows AI to do work, to break down tasks on its own, and to handle processes automatically like an expert.

But after actually using it, I feel it’s not that magical.

﹝Because Skills do not change the capabilities of the model itself﹞

The familiar generation methods are still there; most of the changes happen at the interaction level, not the result level.

In the past, when we interacted with LLMs, we often talked about “tuning,” which means tuning the AI—who it is, how it does things, what to pay attention to, and that it can get scolded if it makes mistakes. In a different scenario, these backgrounds need to be retuned.

﹝Skills move this outside of the conversation﹞

It’s like you pre-write a set of standardized processes that you can use anytime, without needing to restrict the AI’s workflow and methods through chat. It’s essentially a form of engineering-preserved prompts—you could have used prompts to develop this capability from the start, but previously you had to rewrite it repeatedly; now you only need to write it once.

Skills turn these constraints into default states; you only need to tell it what to do at each step.

From this perspective, Skills genuinely improve the interaction experience. You can feel it’s very stable, but don’t expect Skills to suddenly make AI smarter.

For example, I’m currently addicted to vibe coding. I’ve been building my personal system, which allows me to save my most comfortable writing and formatting styles through Skills. Recently, I’ve also been working on a financial system for my mom.

That’s about it. It’s useful, Claude is still the same Claude. But for those who have used others’ Skills, I belike👇