Claude unpacked: commands and skills

I am doing a short email series called Claude Unpacked, where I go through parts of Claude that marketers might not have used yet.

This week, it’s slash commands - in 2 weeks, I’ll cover the best public-created skills.

Aside from the handful of obvious ones, these commands can save your own repeatable prompts, feed Claude your real files, hold your brand rules in place, and keep long pieces of work on track.

Let’s take a look.

** Side note - if you enjoy these newsletters, help me grow and share this link!

A slash command is a shortcut you type with a / to run a saved action. The basic ones you’ve likely seen: /clear wipes the conversation, /help shows the list. Not much to them.

There are two types worth knowing.

1: Built-in commands that ship with Claude and control the session, the model, and the context.

2: Custom commands you write once and reuse, now called Skills (more on that below).

The second type is where the marketing value sits. You take a prompt you keep retyping, save it once, and call it with a single word.

One note on where these run. They mostly live in Claude Code and Cowork, the agent tools, not the standard chat box. So if you’re only ever in the chat window, this is the nudge to open them.

Think of the prompts you type again and again.

Your brief format. Your “five email angles in this structure” ask. A custom command saves that prompt once, so you run it by typing / and a short name you chose.

Making one is simple. The easiest way is to ask Claude Code to set it up: tell it the prompt you want saved and the name you want, and it writes the file in the right place for you.

If you want to know where that is, it’s a hidden folder called .claude inside your user folder. On a Mac that’s /Users/yourname; on Windows it’s C:\Users\yourname. Inside .claude sits a skills folder, and that’s where each command lives. Saved there, the command works everywhere.

You can also keep a command inside a specific project’s folder instead, so it only shows up when you’re working on that project, which is handy for one client or product.

So you’d write your “turn this into a creative brief” prompt, save it as brief, and from then on /brief runs it.

You’ll see these called two things, command and Skill.

A command is the saved prompt on its own, and it runs only when you type its name.

A Skill is the newer version: set up the same way, called the same way, but Claude can also reach for it on its own when a task fits.

Skills are the standard now, so make new ones as Skills.

For marketers, this builds a toolkit out of prompts you already use: /brief, /newsletter-angle, /campaign-audit, /repurpose.

Write a /repurpose command once to say “take this asset and turn it into five email angles, three subject lines, and a LinkedIn post, laid out like this.” Point it at a webinar recap or a landing page and it does exactly that every time, without you re-explaining.

A saved command always does the same thing, because the prompt inside it is fixed. If your /email-angle command has “our pricing page” baked in, it only ever writes about the pricing page.

The fix is to leave a blank you fill in when you run it. That blank is $ARGUMENTS. Inside the file, you write the word $ARGUMENTS where the topic should go:

Write five email angles about $ARGUMENTS.

Then when you run it, you add your topic after the name:

/email-angle pricing page rewrite

Claude drops what you typed into the blank, so it runs “write five email angles about pricing page rewrite.”

Type /email-angle the new onboarding flow next time and it switches topic. One command, any topic, product, persona, or channel, without editing the file again.

Some things you want Claude to know every time, not per task. Your tone. Your positioning. Your audience. The phrases you never want to appear.

/memory opens a file for exactly that. You fill it with those standing facts, the way you’d brief a new hire, and every session starts already knowing them.

No re-explaining your tone, no hoping Claude remembers the “never say this” list. Claude can add notes here as it learns how you work, and you can edit them through the same command.

It’s the home for your style guide and guardrails, so output sounds like you by default.

Claude can only hold so much of a conversation at once. That working memory is the context window, and on a long session it fills up, at which point Claude starts losing what you agreed early on.

/context shows how full it is and what’s taking up room, so you can check before you hit the limit.

/compact clears space by summarising the conversation and carrying on from the summary.

You can steer it: /compact keep the positioning decisions and the three approved angles.

Together they get you through a full strategy block without Claude forgetting the early decisions by the end.

For anything bigger than a quick task, see the approach before Claude starts changing things.

/plan makes Claude lay out how it intends to do the job before doing it.

You read the plan, adjust anything you don’t like, then tell it to go.

Use it before campaign builds, landing page rewrites, email sequences, and site audits, so you don’t discover the approach was wrong after it’s done.

Sometimes you want to test a different angle without losing what you’ve got.

/branch takes a copy of the conversation at your current point and moves you into a separate offshoot.

You explore the new direction there while your original stays untouched, and you can switch back to it whenever you want.

Try a different headline or angle, see how it plays, and keep the version you already liked.

/fork is the version for when you don’t want to stop what you’re doing.

Instead of moving you into the copy, it hands the side task to a background worker that runs while you keep going, then drops its result back into your conversation when it’s done.

So you could fork off “pull three competitor angles for this” and carry on writing while it works.

Start with /branch. Reach for /fork once you want things running in parallel.

On their own each of these helps. Together they run a whole piece of work.

Say you’ve got /memory holding your brand voice and a /repurpose command saved. Turning the Q3 webinar into a launch sequence might run like this:

/plan shows you the approach before anything’s written. /repurpose reads the actual recap file and drafts the sequence in your structure, already in your voice because /memory is in place. /branch lets you test subject lines without touching the approved draft. /compact keeps the session sharp once it runs long. Four commands, one job, nothing re-explained.

You don’t have to build all of it at once. Take one prompt you already retype and turn it into a command. Three ways to do that:

Ask Claude to set it up. In Claude Code or Cowork, describe the command you want and the name, and it writes the file to the right folder for you. No setup on your part.

Use the skill creator. Claude has a built-in /skill-creator (in Claude Code, and in Claude.ai) that walks you through building a skill with a few questions, so you end up with a clean, well-formed command rather than a rough text file.

Borrow one from GitHub. Anthropic publishes a public skills library at github.com/anthropics/skills, and the community has built plenty more. 

Copy one into your skills folder and adapt it to your brand, or install a set straight into Claude Code.

Set up that first command, point /memory at your brand voice and never-say list, and the rest build from there.