I used to drown Claude in setup files. A markdown file for my voice. Another for my rules. One more for banned words. Every new chat meant re-attaching the whole pile, and half the time I forgot one and got generic robot output back. So when I came across this post from an AI professional breaking down a cleaner way to do it, I stopped scrolling.
The creator lays out a simple shift: most of what you keep shoving into .md files should live inside a Claude Skill instead. Skills load automatically in every chat. No attachments. No reminders. Your voice and rules just show up. I think this is one of those small changes that quietly saves you an hour a week, so let me walk you through exactly what the original poster recommends.
Why files were the wrong tool
Here’s the core idea the expert is pushing. Files are for facts Claude can’t guess. Skills are for everything else. A .md file that you have to re-attach and re-explain every session is friction you don’t need. If your voice and your rules never change week to week, they shouldn’t live in a file you babysit. They should live in a Skill that loads on its own.
Files for facts. Skills for everything else.
The step-by-step the creator shared
The whole thing takes a few minutes. Follow it in order and you only set it up once.
- Open Claude Settings, then go to Skills, then Create. This is where the magic lives instead of your file folder.
- Name the Skill something you’ll remember, like /How-I-write. The name is how you’ll think about it later.
- Paste a short template inside and edit the brackets to match you. This is the actual content Claude will pull from every time.
- Save it. Do not attach it anywhere. Close Settings. That’s the whole point, it works in the background.
- Open your old .md files and run each one through one simple test.
- Ask yourself: “Does this need editing or changing every week?” If no, delete it, because it lives in the Skill now. If yes, move it to a folder called Facts Claude can’t guess.
- Open a brand new chat and ask for something you’d normally prompt for.
- Don’t mention the Skill. Don’t attach a single file. Let it run clean.
- Read the output. Your voice, your rules, zero reminders.
- See something off? Fix the Skill once, not the chat. One edit updates every future conversation.
- Repeat for your next workflow and build out your library.
The template to paste in
This is the exact template the post’s author gives you to drop inside the Skill. Copy it as is, then swap the bracketed parts for your own details.
My voice: [how you talk. e.g: short sentences, direct, no fluff]
Banned words: delve, unlock, leverage, [add yours]
My writing samples: [paste your writing pieces]
Rule: Use this on everything unless I say otherwise.
Notice how tight that is. A voice line, a banned words line, a few writing samples so Claude learns your rhythm, and one rule that tells it to apply this everywhere by default. That last line is the quiet hero. It means you never have to remind Claude to sound like you.
Build a library, not a one-off
This is where I got genuinely excited about the approach. The creator doesn’t stop at one Skill. Each workflow you repeat becomes its own Skill, and they all load together in every chat automatically.
- Settings, Skills, Create, then name one /Client-emails for the way you write to clients.
- Settings, Skills, Create, then name one /How-I-design for your design taste and standards.
Stack a few of these and you’ve basically cloned your own working style into Claude. Writing voice, email tone, design rules, all live and loaded, no file juggling. The mind behind this post frames it as building a personal system once and reusing it forever, which lines up with how the best AI users I follow operate.
Tips to make it stick
A few practical notes on top of the original steps, so you get this right the first time.
- Keep facts separate. Client names, project specifics, dates, exact figures. Claude can’t guess those, so keep them in files, not Skills.
- Feed real samples. The writing samples line is doing heavy lifting. Paste actual pieces you wrote, not a description of your style. Claude matches patterns better than adjectives.
- Fix upstream, not downstream. When output drifts, resist tweaking the single chat. Edit the Skill once and every future chat inherits the fix.
- Name skills like commands. The slash-style names keep them easy to reason about when your library grows.
Why this matters
The bigger trend here is worth naming. We’re moving from prompting the same context over and over to setting up a system that remembers it for us. That’s the direction serious AI workflows are heading, and this post is a clean, no-cost example of it. You’re not writing better prompts, you’re removing the need to repeat yourself at all.
The original poster does a great job making a technical idea feel doable in one sitting. If you want the full flow with every detail in their own words, go check out the complete LinkedIn post. Then open Settings and build your first Skill before you talk yourself out of it.