What no one groks about agent skills

You probably have heard that agent skills let you extend what your coding agent can do. And that's true, as far as it goes.

But let me give you a more precise way of thinking about it:

When you use an agent skill, you are DELEGATING A DECISION to the coding agent.

Specifically, you are letting the coding agent decide whether to inject the contents of the SKILL.md file into your prompt, or not.

Because when you install an agent skill, that file is not automatically injected into the context. In fact, you have NO CONTROL over whether it gets injected or not. There is no way to force it, if you want to use agent skills; by design, you do not have that direct control.

And that's a good thing when you use it right.

Because most people don't quite understand how the coding agent decides whether to use a skill.

It is certainly not random; when you create the agent skill properly, you also define metadata that allows the coding agent to decide "do I use this skill or not right now?" And if you set it up right, the coding agent can make a good decision.

This is the difference between vibe coding, versus leveraging coding agents as an engineer.

And frankly, as far as I can tell, most people using coding agents do not understand that mechanism. They cannot tell you how a coding agent decides when to use the agent skill, and when it decides not to. Even though it is obviously important.

This is why it's taught as part of the Agentic Coding Foundations course, one of the key training modules in the Boosted Code curriculum. I mean it when I say completing this course module alone will make you more proficient with coding agents than the majority.

Already, some folks have signed up and are sinking their fangs into it as we speak.

Read about its agentic coding track here: