What an AI skill is, in plain terms
In plain terms, it's just a file that includes the best practices or instructions for how an agent needs to complete one particular task. Every time you ask an agent to do something, it takes that time to think of a way to do it. But if you already know the way you want it to do it, but you just want it to follow that process, you design a skill. And a skill could be anything. It could be: this is how I like to brainstorm, or this is how I like to debug code, or this is our design system.
Instructions, you have to give them all the time. Think of skills as your prompt library, sort of, but then they're a little bit more actionable. It's just a way to make AI more efficient.
If you think about your chat window, your chat window has the system prompt that Anthropic or whatever puts in there, and it has your user prompt. What the skills is, is that it's layered between the system prompt and the user prompt. But it only triggers when it feels like this is something where I need to use this skill for.
One of the skills I've created: when we want to write migration database files, a lot of times AI always hallucinates what tables are in our SQL tables, because it can't see it. Now I have a skill that has our database schema in a markdown file. What I tell the AI: every time you want to write a new file for us, check this, use this skill, because this skill knows how best to write database schemas for our project.
Said out loud on Sensemaking with Wolé & Tobi, Episode 107, “Between Point 'A' and Whatever's Next”, released 21 January 2026. Trimmed for filler; not rewritten.