Sep 22, 2026 · Free Guide

“LEARN” The One Step That Makes Every Claude Skill Get Better Every Time You Use It

If you hate the way Claude writes your website copy, your ad scripts, your emails, or your content, this is why. Every time you open a fresh chat, Claude has forgotten everything you taught it last time. You are training an employee who gets amnesia every night. Here is the fix, and it takes ten minutes to bolt onto any skill you already have.

Why your skills never get better

Most people run their skills like this. Claude writes the script. You read it. You like some parts and hate others. You tweak it yourself, record it, and move on.

That workflow has a hole in it. All your fixes live in the final file. Claude never sees them. Next time you run the skill, it starts from the same blank slate and writes the same stuff you hated last week.

You end up fixing the same three things forever. Too long. Too formal. Wrong opening. Every single time.

The skill is not bad. It just has no memory of you.

Where this comes from

Andrej Karpathy, one of the founders of OpenAI, put a name on this back in May 2025. He called it system prompt learning.

His point was simple. Pretraining gives a model knowledge. Fine-tuning gives it habits. But a lot of how humans actually learn looks like neither. You hit a problem, figure something out, and write yourself a note for next time. "When this happens, do that." Learning by taking notes, not by rewiring your brain.

He argued models should learn the same way. Not by retraining. By editing their own instructions based on what worked.

You do not need to wait for the AI labs to build that. You can do it right now, inside your own Claude Code skills, with a couple of markdown files.

What preference learning actually is

Here is the whole change. One extra step.

Old way
Claude drafts. You edit the file yourself. You ship.
New way
Claude drafts. You give feedback in the chat. Claude revises. You approve a final version in the chat. You ship. Then Claude saves what changed.

That last part is the magic. Because you did the edits in the conversation instead of in a text editor, Claude can see the gap between what it wrote and what you actually wanted. It writes that gap down. Next run, it reads its own notes before it drafts.

That is preference learning. Claude building taste for you, one revision at a time.

The three files that do the work

You do not need code. You need three markdown files living inside the skill folder.

01
approved/
A folder holding every final version you shipped. The last two or three get read before each new draft, so Claude calibrates against your real output, not its idea of you.
02
revisions-log.md
The learning journal. Every time you change a draft, Claude appends an entry. Date, topic, what it wrote, what you changed it to, and the lesson it pulled from the difference.
03
voice-rules.md
The hard rules. When the same lesson shows up in the log three times, it gets promoted here as a permanent rule. Claude reads this file every run. You never give the same feedback a fourth time.

I run this inside my own email skill. The revisions log is now dozens of entries deep. Things like "revenue stat gets followed by the offer shape on the next line" and "hold the price for the release email, not the tease." I said each of those once. Claude wrote them down. It has not messed them up since.

HOW TO ADD IT

How to add it to any skill

You have two options. The paste-it-yourself block, or the prompt that makes Claude do the whole thing and ask you the right questions first. Most people should use the prompt.

Option 1: The prompt RECOMMENDED

Paste this into Claude Code:

Add preference learning to one of my Claude Code skills. Before you touch anything: 1. List every skill I have installed and ask me which ONE to apply this to. Do not pick for me. 2. Tell me if my pick is a bad fit. Preference learning belongs on skills I run repeatedly where my taste decides if the output is good: scripts, emails, ad copy, sales pages, proposals, website copy, DMs, captions. It does NOT belong on research, scraping, data analysis, coding, or anything where accuracy matters more than my style. If I pick one of those, say so and ask again. 3. Ask me if I have any past examples of this type of output that I was happy with. If yes, save them into the approved folder so you have something to calibrate against from day one. Then wire it up inside that skill's folder: - Create voice-rules.md (hard rules, always follow), revisions-log.md (past mistakes, never repeat), and an approved/ folder (every final I ship). - Update the SKILL.md so that BEFORE every draft it reads voice-rules.md, revisions-log.md, and the 2 to 3 most recent files in approved/. - Update the SKILL.md so that AFTER I respond to a draft: if I rewrote it or gave feedback, revise, then append an entry to revisions-log.md with the date, topic, key sections of your draft, key sections of my version, and the specific patterns you extracted from the difference. If the same pattern shows up in 3 or more entries, promote it to voice-rules.md as a permanent rule. If I give a direct rule like "never do X" or "always do Y", add it to voice-rules.md immediately. Save my approved final to approved/YYYY-MM-DD-topic.md. If I approve as-is, save it and note in the log that it hit on the first try. - Add one line to the top of the skill telling me to give feedback in the chat instead of editing the file myself, because edits outside the conversation never get learned. Show me the updated SKILL.md and the three new files before you finish.

Claude will list your skills, make you pick, push back if you picked wrong, and build the whole loop.

Option 2: Paste the block yourself

Already know which skill and comfortable editing files? Open its SKILL.md and paste this at the end, then create the three empty files next to it.

## Preference learning Before drafting: 1. Read ./voice-rules.md (hard rules, always follow) 2. Read ./revisions-log.md (past mistakes, do not repeat) 3. Read the 2-3 most recent files in ./approved/ for tone calibration After the user responds to a draft: - If they rewrote or gave feedback: revise, then append an entry to ./revisions-log.md with the date, topic, key sections of my draft, key sections of their version, and the specific patterns extracted from the difference. - If the same pattern appears in 3+ log entries, promote it to ./voice-rules.md as a hard rule. - If they gave a direct rule ("never do X", "always do Y"), add it to ./voice-rules.md immediately. - Save the approved final version to ./approved/YYYY-MM-DD-topic.md If the user approves the draft as-is, save it to ./approved/ and note in the log that it hit on the first try.

Which skills this belongs on

YES
  • Content scripts, reels, YouTube outlines
  • Emails to your list
  • Ad copy and headlines
  • Sales pages, landing pages, website copy
  • Proposals, DMs, captions, anything a client or audience reads in your voice
NO
  • Research and summarizing skills. You want facts, not your flavor.
  • Scrapers, data pulls, spreadsheets
  • Coding skills. Correct beats "sounds like me."
  • One-off tasks you will never run again. Nothing to learn.

One skill at a time. Get it working on the skill you run most, feel the drafts improve, then add it to the next one.

The one habit that makes it work

This only works if you give feedback in the chat instead of quietly fixing the file yourself.

I know. Editing the file is faster in the moment. But every edit you make outside the conversation is a lesson Claude never learns. You pay for that shortcut every single week after.

So say it out loud. "Cut the first paragraph, too slow." "Never call it a game changer." "Shorter sentences in the hook." Approve the final in the chat. Let Claude do the bookkeeping.

Two weeks in, you will notice the drafts need fewer fixes. A month in, you will approve some on the first pass. That is the compounding most people never get from AI because they never gave it a memory.

All Resources

Karpathy on system prompt learningThe tweet this whole idea comes from
Email Ghostwriter Skill (GitHub)My skill with preference learning built in. Copy the pattern.
Claude CodeYou need this installed to use skills

The Next Step

Guides show you what. AI Builders shows you how.

Inside the community, I walk through every build live, including the stuff that doesn't make it into guides. Regular people (not developers) figuring out AI together, shipping real projects, asking me anything. No fluff, no theory, just the actual work.

Join AI Builders

skool.com/ten-fold