A friend of mine — a project manager — told me she has no use for AI because she's "not really a developer persona." I get it. The AI conversation is dominated by people building things, shipping code, and optimizing pipelines. If that's not your world, it's easy to think AI isn't for you.
But here's the thing: she already offloads work to assistants. She writes briefs, creates checklists, delegates follow-ups. The only difference between delegating to a human and delegating to an AI agent is that the agent needs you to write down what you want.
That's what skills are for.
What Is a Skill?
A skill is a markdown document. That's it. A text file that tells your AI agent how you want something done.
Think of it this way: anytime you'd look at a junior colleague and say "Hey, I need you to do something," they'd say "Great — how would you like me to do that?" A skill is your answer to that question, written down once so you never have to repeat it.
Three Ways to Use a Skill
- Slash command — if your skill is registered, type
/grill-meand Claude finds it automatically - Copy-paste — select all, copy the skill text, paste it into chat, hit enter
- Reference by path — right-click the file, copy the path, pass it to Claude
All three produce the same result. The skill is just instructions. How you deliver them doesn't matter.
The Grill-Me Skill
My favorite skill is grill-me, created by Matt Pocock. It does one thing: forces Claude to ask you questions about your plan before it writes a single line of code.
It's easy to say "I want to build a to-do app." But what about the infrastructure? The deploy model? Multi-user access? The grill-me skill surfaces those questions upfront — the ones you haven't thought about yet — so your agent has the answers it needs before it starts working.
This is the pattern: you don't need to be technical to benefit from this. A project manager running grill-me on a process improvement plan gets the same value as a developer running it on a software architecture. The skill doesn't care what you're planning. It just makes you plan better.
Skills vs. Runbooks
I use two words for the same underlying format:
-
Skills — code-forward instructions. How to interact with an API, how to work with a codebase, how to authenticate with a service. Think: "how do I do something technical?"
-
Runbooks — operational SOPs. First do this, then do that, then do that. Think: "what is my step-by-step process for updating documentation?"
The distinction is mine, not universal. You can call them whatever you want. The point is that some instructions are about how to do a thing (skills) and some are about what sequence of things to do (runbooks). Both are just markdown.
The Real Value: Writing Down What You Do
I hear people say "AI is never as good as when I do it." Of course it's not. But that's not the point.
The point is: if you can write down what you do, AI can get 75-85% of the way there. And after it creates the first draft, you go back and refine. You massage the language. You adjust the output. But you don't start from scratch.
That's the insight my project manager friend was missing. She doesn't need to write code. She needs to write down how she'd brief a new team member on a process. Once that's written down — once it's a skill or a runbook — her AI agent can execute it every time, consistently, without her repeating herself.
The offloading process IS the skill. You're already doing it with humans. Now do it with agents.
What You'll Learn in the Video
- What a skill is and how to write one
- How to use the
/grill-meskill to stress-test your plans - The difference between skills and runbooks
- How to use skills to automate documentation maintenance
- Cross-linking skills, runbooks, and plans to prevent duplication
- Why writing down your process is the highest-leverage thing you can do with AI
