You open Claude, paste in the same wall of instructions you pasted last week: format it like this, never use that word, follow this structure.
You wait for the answer, then do the whole dance again two days later. The instructions live in a notes app, or worse, in your head.
There's a fix for that, and it takes about ten minutes to set up: a Skill.
Build one and Claude already knows how to perform the task (exactly the way you like it) before you ask.
A Skill is a saved recipe Claude follows on its own. You write down how you want something done once, and from then on Claude reaches for those instructions automatically whenever the task comes up.
You taught it the rules; now it just behaves that way.
Anthropic shipped these to everyone, and they're the most useful, practical AI feature that you can set up in an afternoon.
Today, I’m walking you through how (and why) to build your first Skill 👇
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
Why a Skill beats a really good prompt
A great prompt is only a one-time instruction.
You type it, Claude does the thing, and the instruction evaporates when the chat ends. Next time, you have to start from scratch.
A Skill is that same instruction, written down and left on the shelf where Claude can grab it.
The difference comes down to three things:
It triggers itself. When you start a new chat and ask for something the Skill covers, Claude recognizes the job and loads the rules on its own. You don't paste anything or even mention the Skill exists. You just say "clean up these notes" and it already knows what you mean.
It stays consistent. A prompt you retype from memory drifts. You drop a rule here, soften one there. A Skill is the same every time, so the output is too. That matters when the format is the point: a report your boss expects in a particular shape, a summary your team reads the same way every week.
It can carry more than text. A Skill is really a little folder. The core is one file of plain-English instructions, but it can also hold examples, a template, even a small script. Claude pulls in whatever the task needs. (More on the folder in a second, you won't have to touch any of it for the build below.)
Behind the scenes it's simple: at the start of every chat, Claude reads a one-line summary of each Skill you've got (just enough to know when each one applies).
It only opens the full instructions when the moment is right. So you can stack up a dozen Skills without slowing anything down or cluttering your chats.
They sit in the background until they're needed 🤝
It’s like the difference between explaining your coffee order every morning versus a barista who already knows it.
Build your first skill
Let's make something you'll actually use.
A lot of people leave a meeting with a mess of notes (half-sentences, arrows, names with no context, three action items buried in a wall of text).
A notes-cleaning Skill turns that mess into the same tidy summary every single time.
You don't need to write any code or hunt through menus.
Claude has a built-in Skill for building Skills (it sounds meta, but stick with me).
To get started, you just ask it to walk you through building a skill. It interviews you and writes the whole thing up.
Step 1- Kick it off. Open a chat with Claude and paste this:
Let's create a skill together using your skill-creator skill. First ask me what the skill should do.
Step 2 - Tell it what you want. When Claude asks what the Skill should do, describe your meeting-notes cleaner:
Clean up my raw meeting notes. When I paste in messy notes, turn them into a summary with three sections: a two-sentence recap at the top, a bulleted list of decisions we made, and a list of action items that each name who owns it and the deadline if one was mentioned. Keep it short, and don't invent action items that weren't actually discussed.
Step 3 - Answer the questions. Claude will ask you the things you didn't think to specify. Stuff like:
Should it flag items where no owner was named?
Do you want the recap in past tense?
What should it do with side tangents?
Answer in plain language. This back-and-forth is where the Skill gets good. You're handing over the judgment calls you'd normally make in your head.
Step 4 - Save it. When Claude shows you the finished Skill, save it to your Skills (you can ask it to do this for you). Done. It's now on the shelf.
Step 5 - Use it without thinking about it. Open a fresh chat tomorrow, paste your raw notes, and type "clean these up." Claude recognizes the job, loads your rules, and hands back the exact format you designed (without you mentioning the Skill at all!). That's the whole payoff: you taught it once, and now it's just how Claude behaves.
The first version won't be perfect, and that's fine.
Use it for a week, notice what bugs you, then ask Claude to tweak the Skill.
They're meant to be tuned over time!
Alright, use the below to turn Claude into a Skill-writing partner for anything you do repeatedly.
Paste it into a fresh chat and swap in your own task 👇
Let's create a skill together using your skill-creator skill. It should handle [describe the repeated task — e.g., "writing first-draft replies to customer support emails in our company's tone"].
Interview me before you write anything.
Ask me about:
- exactly what the finished output should look like
- the rules and preferences I'd normally keep in my head
- words, phrases, or formats to always use or always avoid
- what you should do when you're unsure or information is missing
Ask your questions a few at a time so it stays a conversation, not a form. When you have enough, write the Skill and explain in one line when it'll trigger.
A good first candidate is any task where you've ever thought "I wish it would just remember how I like this."
Drafting in your voice, formatting a weekly update, summarizing a document type you see constantly, turning rough ideas into a standard brief.
If you do it more than twice a month, it's a Skill.
Your roundup of the latest model releases and updates from the biggest AI labs.
Meta is building a paid AI agent called Hatch. Reports say Meta's first paid AI product is an agent — software that takes actions for you, like booking appointments, sending emails, or shopping online — with a free tier and a "Hatch Plus" plan rumored as high as $200 a month. Internal testing is wrapping up by the end of June, with a broader US launch eyed for July. Worth noting for builders: Hatch reportedly runs on Anthropic's Claude during development before Meta swaps in its own model — a sign of how much the big platforms still lean on each other's engines. (The Decoder)
Claude Mythos could go public today . Anthropic's most capable model, Mythos, has been locked to security partners until now. The public version is rumored to land this week (with some outlets reporting that it will launch today) with more guardrails and lighter cybersecurity functions than the partner build, but a real jump in handling long, multi-step jobs. If you've been waiting to test it, this is the week. (The Information)
OpenAI and Anthropic both filed to go public. OpenAI submitted confidential IPO paperwork on June 8, a week after rival Anthropic did the same — the two biggest names in AI now lining up for the stock market, days ahead of SpaceX's expected trillion-dollar debut. OpenAI was last valued near $852B (after a $122B raise in March), Anthropic around $952B. OpenAI says the timing isn't set — "it may be a while" — but the back-to-back filings signal that the companies building these tools expect serious, durable demand. Worth noting if you're deciding which tools to invest your own time in. (NBC News)
Advertise with Build with AI
Get in front of an audience of professionals using AI day-to-day: founders, engineers, operators, and product builders.
Interested in advertising? Respond to this email for rates and details.
Until next time,
William Ryan
Editor-in-Chief @ Build with AI
PS: Follow me on X for daily updates and AI workflows.