When I first got into Claude Skills, I watched a LOT of videos made by incredibly clever people saying incredibly clever things. And my brain just sort of shut down… because they were about thirty steps ahead of where I was.
So this is the version I wish I’d found at the start: five Claude Skills built around your business, and the one that runs them.
Jump to a bit of the video:
1:22 Your first skill: teach Claude your business
3:24 Skill one: the idea finder, built from your saves
5:31 Do you need a paid Claude plan?
6:46 Skill two: the hook writer, and why it stands alone
7:35 A carousel hook is two slides, not one
9:03 Skill three: the carousel writer in your own Canva template
12:17 Skill four: the caption writer, complement never recap
13:44 Skill five: the filer and your content calendar
14:35 The “before it goes live” column
16:07 My biggest mistake: one skill that did everything
16:34 One master skill that runs the other five
17:04 Why I don’t automate the whole thing
Want all seven prompts from the video? Get the free Claude Skills prompts here. Each one is ready to paste into Claude, and each one interviews you before it builds anything.
The short version
The five Claude Skills worth building for your social media, once Claude knows your business:
- An idea finder that turns your Instagram saves into post ideas.
- A hook writer that knows a reel hook from a carousel hook.
- A carousel writer that fills your own Canva template.
- A caption writer that complements the post and never recaps it.
- A filer that puts everything in your content calendar.
Build a brand skill first so every skill knows who you are, then one master skill that runs the five in order and waits for you between steps. Guided, not automated.
What is a Claude Skill, in plain English?
A Claude Skill is a saved set of instructions Claude follows whenever you ask it to do a particular job, so you explain that job once instead of in every chat. For social media, that means one skill per repeatable task: finding ideas, writing hooks, building carousels, writing captions and filing the finished post.
They can get really complicated really quickly, which is the bit that lost me at the start. So everything in this post is the beginner-friendly version.
Full disclosure, though: we’re doing this the hard way. You could grab a generic skill I made for my business and plop it into yours… but it would know my audience, my pain points and my voice, not yours. So each of the seven prompts builds the skill by interviewing you first. It takes a little more work up front, and what comes back is yours, not mine.
What should you set up before your first skill?
Set up a brand skill first. It holds everything about your business, your audience and your voice in one place, and every other skill reads it. If you already have a project or a file in Claude that knows all of that, you can skip this step.
Because here’s the thing: if you build a caption writing skill and it doesn’t know who it’s writing captions for, it’s not going to do a very good job.
It doesn’t write anything at first. It interviews you, and it pushes back when you’re vague. Say “my audience are business owners” and it will tell you that’s not specific enough at all, then keep asking until it is. It covers:
- What your business does and who you serve.
- What those people are worried about, in their actual words.
- Words you never want to use, and words you definitely do.
- Your fonts, colours, website and socials.
- Claims you never want to make.
- Five things you’ve written that sound like you.
Are five writing samples enough? Probably not… but they’re enough to start with, and you can keep adding to them. If you want to go deeper on that, I wrote a whole post on how to make AI sound like you, which pairs with a separate video on exactly this.
Once it’s built, every skill after it can reference it. So you’re not answering the same questions five, six, TEN times every time you build something new.
How do you turn your Instagram saves into post ideas?
Build an idea finder skill that reads one Instagram saved collection and hands back post ideas in the formats you actually make, each linked to the post that sparked it. It works because it uses something you’re probably already in the habit of doing. Instagram is what I use, but it works with wherever you save ideas.
Like the brand skill, it interviews you first:
- How many ideas you want, based on what you can realistically make. There’s no point getting twenty ideas if you plan on making two reels that month.
- Which formats. Reel ideas are useless if you have no intention of making reels.
- How far back to look. In a fast-moving industry like mine, ideas from six months ago aren’t relevant. Sixty days is about my maximum, but it depends on your industry.
- Which collection. It lists your collections and asks you to pick one.
That last question matters more than it sounds. If your saves are full of recipes, dog videos and hair tutorials (mine are), none of that makes good content for your business.
One limit worth knowing: it can only work with what you saved. If you keep saving from the same two people, it will keep telling you their ideas are great when they might not be. So save from a wider pool.
There are much more complicated setups that pull competitor and industry data without you saving anything, with APIs and all the techie bits… this one is the starting point, not the ceiling.
Do you need a paid Claude plan for this?
For the idea finder to read Instagram by itself, yes. That needs the Claude in Chrome extension, which is on paid Claude plans only (checked October 2026).
The free workaround is screenshots and URLs, which is totally solid and just takes a bit longer. And the content calendar doesn’t need a paid plan of any kind: I use Notion’s free plan, and it works fine.
Why should hooks be their own skill?
Hooks should be their own skill because getting someone’s attention quickly is a bit scientific, and it works completely differently depending on the content type. Skills that bundle hooks in with captions or scripts end up super messy.
So the skill starts by asking which kind of hook you need:
- Long-form video, like a YouTube video. You have about 30 seconds.
- A reel. You have about three seconds.
- A carousel. This is the one I kept getting wrong.
- A written caption, like a LinkedIn post or an Instagram or Facebook caption. People only see a little text before “see more”, so the hook has to earn the tap.
A carousel hook is two slides, not one. The first slide stops them; the second slide earns the swipe. My mistake was saying the same thing on slide two that I’d already said on slide one, which gives nobody any reason to keep going.
Then it writes four VERY different options and tells you which one it would use, and why. The final choice is still yours; it just gives you something to bounce off, like a proper brainstorming partner.
Can Claude build a carousel in your own Canva template?
Yes. A carousel writer skill can take the link to your branded Canva template, read how it’s laid out, write the slide copy and then put that copy into your design. The template stays exactly as you made it, because the skill only supplies the words.
I’m using carousels here, but the same approach works for other content types. A few things to set up before you start:
- A branded Canva template, ready before you begin.
- Spare slides. Even if you’re 100% sure you’ll only ever post ten, keep twelve or thirteen in the template. Claude can delete the extras instead of inventing new designs from scratch.
- Image folders Claude can pull from, so it’s not guessing what you want. You can always swap them out.
It then analyses your design. Mine has a heading and a body; yours might have your handle or page numbers too, and it works out how to write for that. If the idea could use personal touches, it interviews you for your stories, your numbers and your ideas, so the carousel isn’t generic advice anyone could have written.
And it lays out the copy as plain text before anything goes into a design. That step is SO important, because editing words on a page is much easier than editing them once they’re designed.
The first version won’t be exactly how you’d write it. So rewrite it, give your version back to Claude, and tell it to learn from that and make the skill better. Otherwise it has no idea that what it gave you and what went out on Instagram were two different things.
On mine, I still tweaked some spacing and swapped out the first image. It’s never going to be 100%… but it takes a job that might have taken an hour or an hour and a half down to maybe five or ten minutes.
What’s the one rule for an AI caption writer?
The caption complements the post. It never recaps it. If someone can read your caption and skip the carousel, you’ve done yourself a massive disservice. It’s a rule I constantly broke… right up until I actually wrote it down.
So the caption writer reads what you’ve made first. That can be the finished carousel or just the slide copy. Two smaller rules sit inside it:
- Never repeat the first slide. If someone is looking at slide one and the caption underneath says the exact same thing, that’s a bad experience for them.
- Never invent a second call to action. If the carousel already asks people to like, save or comment a keyword for an automation, the caption respects that and doesn’t ask for something else.
That second one is something AI gets wrong a lot. The carousel asks for one thing, the caption asks for another, and the two don’t gel at all.
Where should finished content live?
Finished content should live in your content calendar, not in an AI chat, so the last skill is a filer. It reads your existing calendar, or helps you build one, then files the post with everything you need to publish it. If it’s sitting in a random chat and you close it or forget about it, it’s never going to go live.
I use Notion, and I don’t pay for it. It still works, so you’re not limited by a paid plan here. You don’t have to use Notion at all, though; use whatever works for you.
If you already have a calendar, it reads your columns and categories and builds the filing around them. It files the Canva link to the finished design, the caption, and any optimisations, like a location, a product or a person to tag.
I also added a “before it goes live” column. If a post uses a keyword automation, that automation has to exist before the post goes live, and it’s the bit that gets left behind. Picture a post happily going viral, a thousand people commenting the keyword, and nothing arriving in their DMs. That column is how I prevent it.
Then the filer stops… it doesn’t do anything else. Could it? Absolutely. There are lots of schedulers that plug into Claude and will schedule your content once you’ve approved it.
I’ve left that out on purpose: I like having my hand in the whole process, and it’s worth fine-tuning every skill before you let one publish for you. If you have a scheduler set up in Claude, you can build that step in later.
Can one skill do everything?
Not well. Build five skills that each do one job really, really well, then one master skill that runs them in order and waits for you between steps. That’s the fix for the biggest mistake I made.
My mistake was a YouTube skill that did my research, my thumbnail planning, my scripting and my optimisation all in one go. It was a mess. If a skill is doing everything, it’s probably not doing any one thing very well.
Setting up the master skill is a lot less technical than you might think. You describe the sequence in plain words and Claude writes the skill. Something like:
I want a skill that runs my idea finder, then waits for me to choose, then runs my hook writer on the ones I’ve picked, waits again, then runs my carousel writer, and so on.
This is NOT a fully automated process. It guides you through step by step and cuts the thinking you have to do in half, or less than half.
Could you automate the whole thing? Yes. Could the other end also look like complete AI slop? Also probably yes. So it really depends on what you want to come out the other end.
Go and build something, even if it’s just one of these skills. Start with the brand skill, then whichever one would save you the most time this week. And if there’s a job you can do but can’t quite describe, you can skip the writing and record a Claude skill while it watches you work.
If you’d rather not start from a blank chat, all seven Claude Skills prompts from the video are free, ready to paste into Claude.
Originally published October 2026.
