How I Add Skills to Claude — and Why My Workflow Will Never Be the Same

There was a specific moment a generic assistant became genuinely useful to me, and it wasn't a bigger model or a cleverer prompt. It was the first time I figured out how to add skills to Claude. Before that, I was re-explaining the same context every session — my file structure, my conventions, the way I like things done. After that, Claude just knew, and the whole thing stopped feeling like babysitting and started feeling like delegation.

I want to be honest about this, because the internet is full of breathless "10x your output" takes and most of them are noise. Skills are real, they're useful, and they're also narrower than the hype suggests. Here's what they actually are, how you add one, and where they helped me — and where they didn't.

What are Claude Skills?

A skill is a folder of instructions and (optionally) scripts that Claude loads on demand when a task matches. Think of it as a reusable playbook: instead of pasting the same three paragraphs of "here's how we do things" into every conversation, you write it down once, and Claude pulls it in automatically when it's relevant.

The mechanics are refreshingly boring. A Claude skill is a directory with a SKILL.md file at its root. That file has a little bit of metadata at the top — a name and a description — and then a body of instructions written in plain Markdown. The description is the important part: it's what Claude reads to decide whether this skill is relevant right now. If the description matches what you're asking for, the full instructions get loaded into context. If it doesn't, they stay out of the way. That "load only when relevant" behavior is the whole trick — it's why you can have dozens of skills installed without drowning every conversation in irrelevant instructions.

For anyone using Claude Code — the CLI — skills are how you encode your project's rules once and stop repeating them. A skill can hold your commit conventions, your testing checklist, the way your monorepo is laid out, whatever. The point is the same across surfaces: turn tribal knowledge into something the model reaches for on its own.

How do you add a skill to Claude?

Here's the part people overcomplicate. Adding a skill is mostly making a folder.

1. Create the directory. Skills live in a skills/ folder — for Claude Code that's typically ~/.claude/skills/<skill-name>/ for personal ones, or a project-level .claude/skills/ directory you check into the repo so your whole team gets it.
2. Write SKILL.md. At the top, a short frontmatter block with a name and a description. Then the body: the actual instructions, written the way you'd brief a competent new teammate.
3. Nail the description. This is where I spent the most time and where it mattered most. The description is the trigger. "Helps with code" is useless — it either fires constantly or never. "Use when creating or editing App Store screenshot layouts, including sizing and localization" is specific enough that Claude knows exactly when to reach for it.
4. Reference supporting files if you need them. A skill can point to scripts, templates, or reference docs sitting alongside SKILL.md. Claude only pulls those in when the task actually needs them, so you can keep a heavy reference doc out of context until it's relevant.
5. Use it. Start a conversation, ask for something that matches, and watch it load. That's it — no build step, no install command, no restart ritual.

The first time I did this it felt anticlimactic, like I'd missed a step. I hadn't. A skill is just a well-written document in a folder Claude knows to look in. The craft is entirely in what you write, not in any tooling.

Which skills actually changed my workflow

Not every skill earns its keep. A few genuinely did:

• A project-conventions skill. One SKILL.md describing my repo layout, my "logic goes in the lib package, never inline in a component" rule, and my commit style. I stopped correcting the same mistakes every session. This one alone justified the whole exercise.
• A "before you commit" checklist skill. Run the typecheck, run the linter, don't add an AI co-author trailer. Small, dumb, and it saved me from shipping broken commits more than once.
• A domain-language skill. The vocabulary and rules specific to what I'm building, so Claude stops guessing what my terms mean and starts using them correctly.

The pattern across all three: the skills that changed my workflow weren't clever. They were the boring, repeated context I was tired of typing. If you find yourself explaining the same thing to Claude a third time, that's your next skill.

This is the same workflow instinct I lean on outside of coding. When I ship App Store creative for my apps, I don't want to hand-run the production line every time either — I paste a listing URL into Reverze and let it rebuild my screenshots into an editable, every-size campaign, the same way a skill lets Claude rebuild my context without me re-briefing it. Encode the repeatable part once; spend your attention on the part that actually needs you.

What skills can't do

Here's the honest ceiling, because you should know it before you get evangelical.

A skill won't make Claude smarter. It doesn't add reasoning ability or new knowledge — it adds instructions. If the underlying model can't do something, wrapping it in a skill won't change that. Skills shape behavior; they don't raise the ceiling.

They also don't fire reliably if your description is vague. A badly-scoped skill either never triggers (and you wonder why nothing happened) or triggers on everything (and pollutes context). Half of "learning skills" is really learning to write descriptions that mean something.

And they're not a substitute for judgment. A skill can hold my checklist, but it can't decide whether the thing I'm building is a good idea. The taste, the "should we even do this" call — that stays mine. Which is fine. That's the part worth keeping.

Read next

If you want the version of this that goes deeper on using skills to level up the quality of what you ship — not just the speed — I wrote about that in how I improved my design sense with Claude skills.

The short version of everything above: skills are how you stop repeating yourself to an AI. They're not magic and they're not a bigger brain. They're a folder with a good document in it — and that turned out to be enough to change how I work every single day.

---

Reverze turns App Store creative production from a multi-day design workflow into minutes — paste a URL, rebuild your screenshots, generate new directions, and export production-ready assets. Explore the free tools or start in the app.