First, what is a Skill?
Many people say a Skill is just a collection of prompts—a way to store your prompts to prevent repeated typing. That’s true, but it’s not the whole picture.
The “true” part is that for some very basic Skills, a single Markdown file might be enough. But that doesn’t cover everything. You can include scripts, combine one Skill with others, or reuse and modify them.
Does this sound familiar to the programmers out there?
Yes, it’s similar to the concept of functions. A function can be simple, like a + b, or it can build more complex logic on top of that, like formatting output. In the era of “vibe coding,” prompt is code, so a collection of prompts is also the foundation of automation.
The difference between automation via Skills and traditional workflows
Traditional workflows often act like state machines. For example, you hard-code exactly what to do when a value is >0 or <0. There is no room for ambiguity. If the output has issues, the entire workflow crashes, but as long as the output is correct, the results are guaranteed to be consistent.
AI using Skills has a lot of “fuzzy space.” As long as the output isn’t completely off track, you will likely get a result. You don’t even need tight integration; smart LLMs can connect the dots themselves.
However, the output isn’t always stable. For example, if you build a document-organizing Skill but don’t specify strict naming conventions in your prompt, the file styles might differ every time.
So, how do you get started?
First, clarify your needs. Do you have a need for automation? specifically, automation that allows for some ambiguity. If you already have a stable workflow or don’t need automation, there’s no need to force it.
Although there is a lot of new tech online and models are iterating fast, don’t be anxious. If you have no need, just ignore it.
You have a need, but don’t know how to write it?
It’s simple: State your requirements and let the AI write it.
There are many tutorials online claiming AI-written code isn’t good or efficient enough. My advice is: get from 0 to 1 first, then see if you are satisfied with the result.
If it works, it’s OK. We want results, not process. As long as the AI-written Skill achieves the result you want, you don’t need to change it.
Only when you feel the need to improve efficiency, reduce costs, or handle more complex tasks should you look into it deeper. If there is no need, don’t be anxious, and don’t invent needs out of thin air.
My advice on what NOT to do
I don’t recommend using Skills you don’t understand. Skills can contain Trojans and backdoors. If you use external Skills, you must let the AI or yourself audit them to ensure safety before using them.
It is best to understand the mechanics. If you don’t understand the mechanism, how can you guarantee the result will satisfy you?
I don’t recommend buying Skills if you don’t understand them. Leaving security risks aside for a moment—this might be a hot take—but if you are rushing to spend money on something you don’t understand, the risk of information asymmetry is high, and you are likely to get ripped off.
Even though “prompt is code,” doesn’t the business of buying code when you don’t understand code sound very risky to you?
One final sentence: Be driven by needs. Don’t pay for things you won’t use.