Give Me 13 Minutes and I'll Make You Disgustingly Good With Claude
A former agency art director walks through six moves that take you from chatting with Claude to handing it real work: context files, plan mode, the seven-part goal prompt, skills, and feedback loops.
Posted
yesterday
Duration
Format
Tutorial
educational
Views
28.5K
453 likes
57 · 43
Big Idea
The argument in one line.
Claude Code gives generic results when you brief it with one sentence, and gives work that sounds like you when you give it context files, a plan, a seven-part goal prompt, and reusable skills.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You use Claude mostly in the chat window, get generic answers back, and keep re-explaining your voice and preferences every session.
You're a non-coder, creator or consultant who's heard of Claude Code and is put off by the word code.
You run repeatable jobs each week, like updates, posts or page checks, and want them to run from one command.
SKIP IF…
You already run a CLAUDE.md index, goal prompts and your own skills library and want advanced agent orchestration.
You want a code-level tutorial on building software with Claude Code. This is a workflow-habits video for non-developers.
TL;DR
The full version, fast.
Chat Claude can't do much for you, but Claude Code can make files, build sites and publish work, and you can just talk to it like a new hire. Rogoff's six moves: choose between an SOP (exact steps) and an end goal, split your context into small markdown files with a slim CLAUDE.md as the index, plan before building, brief with a seven-part goal prompt that ends with work autonomously, turn a job that finally worked into a skill, and refine with feedback in small steps so each fix is saved back into the skill.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
The hook, credentials, and why the Code tab is for non-coders too.
01:26 – 02:22
02 · SOPs versus end goals
Two kinds of instructions: exact steps or a clear goal.
02:22 – 04:13
03 · Give it a memory
Split context into small markdown files with CLAUDE.md as an index.
04:13 – 05:42
04 · Plan before you build
Plan mode or just asking for a plan, with the performance marketing platform example.
05:42 – 07:40
05 · The seven-part goal prompt
The seven parts, the milk website comparison, and the free goal skill.
07:40 – 10:31
06 · Turn it into a skill
Describe the job up front, or convert a chat that worked into a skill.
10:31 – 12:31
07 · Refine and iterate
Give feedback in small steps and save every lesson back into the skill.
Atomic Insights
Lines worth screenshotting.
Most people only ever give Claude an end goal in one sentence, then wonder why the result is generic.
A standard operating procedure has largely replaced no-code automation for the author, because Claude can follow written steps from a contact form to a CRM to a branched email.
Packing everything about you into one CLAUDE.md causes drift and made-up details. Keep it as a slim index that points to separate voice, brand, audience, offer and story files.
The most important rule for context notes is that notes point to facts and never copy them. A fact duplicated in five notes will eventually go stale in one, and Claude will confidently use the wrong one.
Plan mode is really for the human, not for Claude. Typing make a plan first gets most of the same benefit.
Planning a project takes 10 to 20 minutes, while building without a plan can cost hours or days.
A goal prompt has seven parts: what you want and why, the quality bar, tools, creative freedom, self-checking at least three times, delivery, and a one-sentence restated goal.
End a goal prompt with: work completely autonomously and do not ask me for anything until you're all done.
You don't write the goal prompt yourself. Ramble to Claude for 5 to 15 minutes and let a skill organize it into the seven parts.
If you'd explain a job to a new hire the same way every time, it should be a skill.
A skill needs a when clause at the top. Without it the skill works fine but Claude never knows to use it.
The step you always forget is usually the most important line in the skill.
Naming the exact tool in the skill, like yt-dlp for YouTube transcripts, stops Claude wasting tokens guessing at routes.
The better way to make a skill is after the fact: when a chat finally gets something right, tell Claude to turn what you just did into a skill.
A skill must be a folder containing a file named skill.md. A loose file is not read, and you need a new session before it appears in the slash list.
Asking Claude to edit a whole video in one prompt fails because it doesn't know what a mistake is or what design you want.
Define what good looks like, give a polished reference, then work one piece at a time: script, design, storyboard, animation.
Takeaway
Brief it like a new hire, then bottle it
WHAT TO LEARN
Claude gives generic results to one-sentence briefs and personal results to a system of context files, a plan, a structured goal prompt and saved skills.
Decide first whether you are giving an SOP with exact steps or an end goal. Most people give only a one-sentence goal and get generic output.
Split context into small markdown files for voice, brand, audience, offer and story, and keep CLAUDE.md as a slim index so Claude only reads what a task needs.
Keep each fact in one place and point to it from other notes, so a stale duplicate never overrides the current version.
Plan before building, by plan mode or just asking for a plan. Ten to twenty minutes of planning avoids hours of rework.
Write goal prompts with seven parts, and let a skill turn a rambled brain dump into that structure instead of writing it by hand.
Turn any job you explain the same way every time into a skill, with a clear when clause and the step you always forget.
After a chat finally works, tell Claude to turn it into a skill. Make sure it is a folder with a skill.md file and start a new session.
Refine in small steps with clear references and feedback, and save each lesson back into the skill so you only give a note once.
Glossary
Terms worth knowing.
SOP
Standard operating procedure. A written, step-by-step set of instructions you give Claude when you want a specific path followed, such as handling a contact form submission.
CLAUDE.md
A markdown file Claude Code reads at the start of every session. Best used as a short index that points to other files rather than holding everything.
Plan mode
A Claude Code setting where Claude lays out what it will do and waits for your approval before building anything.
Goal prompt
A long-form brief with seven parts that tells Claude what you want, what good looks like, which tools to use, how much freedom it has, how to check its work and where to deliver.
Skill
A folder with a skill.md file that packages a repeatable job so Claude can run it from a slash command.
Markdown
A plain text file format, with the .md extension, that Claude reads easily.
“Claude Code doesn't have to mean you're coding, and you don't need to know how. You can just talk to it like you would a new hire.”
removes the main fear for non-coders→ TikTok hook↗ Tweet quote
03:35
“Notes should point to facts and never copy them.”
one tight rule, no setup needed→ newsletter pull-quote↗ Tweet quote
08:41
“That step you always forget usually ends up being the most important line in the skill.”
counterintuitive and specific→ IG reel cold open↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
If you've been using Claude to chat, you're getting a tiny fraction of what it can actually do for you. I'll take you from typing questions into a chat box to having Claude do real work for you, like a pro. I've never written a single line of code by hand, and today I do most of my work in Claude Code.
I'm a former art director for brands like Apple, PlayStation, and Nissan, and I now run one of the top communities for learning Claude Code with over 8 ,000 members. And from teaching all of those people, I know that the ones getting generic stuff back and the ones who get work they put their name on are doing - a handful of things very differently get those wrong and you're re -explaining yourself every time and fixing whatever it hands back get them right and a job that used to take me an hour or two now runs from one command while i do something else so in this video i'm going to go through the moves that took me from chatting with claude to doing my real work with it so you leave disgustingly good with claude and the work it gives you actually looks and sounds like yours it starts with the tool you're using most people start with claude in the chat window you ask it a question it gives you an answer, and that's pretty much where it ends.
Chat is great. You can chat all you want in regular Claude, but it can't actually do very much for you. Claude Code can do everything chat can do and way more.
It can make the files, build the website, write the post, and put it where it needs to go. If you've got the Claude desktop app, it's already in there. Just click the code tab at the top, and that's Claude Code.
I know the word scares a lot of people off, but Claude Code doesn't have to mean you're coding, and you don't need to know how. You can just talk to it. like you would a new hire and just like a new hire it works way better when you give it clear instructions there are really two kinds of instructions you can give it the first is an sop or standard operating procedure that's for when there's a very specific outcome or a specific path you wanted to take say somebody fills out the contact form on your website you take their contact information you add it to your crm which is just the place you keep your customer list then based on their answers to a few qualifying questions you either send them one type of email if they're qualified or or a different type of email if they're not.
Honestly, SOPs have sort of replaced no -code automation for me. The second kind is an end goal. You tell it where you want to end up, and it works toward that based on its own knowledge.
Most people only ever give it the second kind with a single sentence, and then they wonder why it comes back generic. So the first move is to know which one you're giving it. If you have a way you like it done, spell out the steps.
If you don't, give it a really clear goal, and I'll show you exactly how to write one in a few minutes. But before any of that, Claude needs to know who it's working for. The most common problem I see is people who feel like Claude doesn't know them.
tone of voice or how they like to write. They're constantly re -explaining themselves. It's often because they've tried to pack way too much information into one file.
Claude Code reads a file called ClaudeMD at the start of every session. The MD just means markdown, which is basically a plain text file. ClaudeMD is really just the start.
What helps a ton is a handful of other markdown files that each cover one part of you. That's things like your tone of voice, your brand guidelines for visuals, who your audience is and the language they use, your core offer, and your personal story.
Then you strip ClaudeMD way down and use it like an index. It's just a map of where everything else lives. So when Claude writes something in my name, it goes and grabs my voice file.
When it designs something, it grabs the right design file. When it doesn't need either one, it leaves them alone. The whole idea is that Claude only reads the right information when it needs it and nothing more.
That way you're not adding bloat or extra context it doesn't need, which is what leads to it drifting or making things up. Let me show you mine. I've got a section on where all my notes live.
One on my voice, one on my design styles, and a list of files it should only read when it needs them. There's even a section on how I communicate. I use voice to text so it knows to read for what I mean.
The most important rule in my whole file is that notes should point to facts and never copy them. If the same fact lives in five different notes, sooner or later one of them is out to date. Then Claude goes with the wrong one and it's totally confident about it.
So keep each fact in one place and point to it from everywhere else. This is what I call your Claude knows me card. Once it's set up, it just changes the quality of what you get back.
People start feeling like they're able to post more often or that their websites reflect them better without having to re -explain. You can do things on repeat with the confidence that it's all going to be consistent and on brand. Once Claude knows who you are, the next move is getting it to plan before it builds anything.
Claude Code has something called plan mode. You turn it on and instead of jumping straight in, Claude lays out what it's going to do and waits for you to approve it. I use it for any task that has three or more steps.
Now, here's the funny thing. Boris Cherny, who created Cloud Code, has said that you don't really need plan mode anymore. The models are so good, they'll make a plan on their own, or you can just ask for one.
And he's right about Cloud, but I find it's still really helpful if you're new. You get to see the plan of attack before it starts building. So plan mode is really for the human, not for Cloud.
If you'd rather skip plan mode, just type make a plan first, and you get most of the same benefit. When people skip it, they just try to build the whole thing in one. shot.
Then they wonder why it doesn't work. When you plan first and break it into smaller sections, you get way better results. You also save tokens because you're not rechecking a whole system on every message.
We built a whole performance marketing platform for a big financial company, loads of moving pieces, loads of stakeholders, loads of departments to serve. Planning it out first saved us a massive headache. We could see the whole product before we actually built anything and work out how to build it one piece at a time.
Without that plan, we'd have ended up with a bunch of parts that didn't fit together with a lot of bloated repetitive code. And planning is quick.
You can make and refine a plan in 10 to 20 minutes. Building without a plan could take hours or days because you haven't thought through the details. So you've got a plan and the next thing is giving Claude a real brief and this is where most people get stuck.
People get flustered when they give it a task that's too big and they're unhappy with the results. Or they think they're going to give it one prompt and it's going to get them exactly what they're looking for. So what fix is that is a goal prompt.
Every goal prompt I write has seven parts. Part one is what you want and why it matters. Say the real thing you need and the real number.
Part two is the quality bar. That's what excellent actually looks like described in really vivid words. Part three is the tools.
You name the tools that already exist and then you tell Claude to go find more. Part four is creative freedom. Claude is allowed to change the plan and Claude gets to make the choices.
Part five is checking its work. It goes back over what it made. at least three times.
Part six is delivery. That's where the finished thing goes and what link or file comes back to you. Part seven is one sentence that says the goal again, and then this.
Work completely autonomously and do not ask me for anything until you're all done. The idea behind all seven is to get out of the model's way. Now, you don't have to write all seven yourself.
I built this into a skill. You just ramble to Claude for five, 10, 15 minutes about what you have, where you want to go, and any of your ideas. Claude or - organizes it for you into those seven parts.
Let me show you the difference. I'll do the same job two ways. The first way is how most people do it.
Make me a website all about the wonders of milk. So that's what comes back. It works, but it's pretty generic.
For the second one, I rambled to Claude about what I actually wanted. I gave it a beautiful reference and said, this is what I want to look like. Then the goal skill turned that into a full goal prompt.
Here they are side by side. It's the same idea and the same Claude. The only thing I changed was how I briefed it and that's the difference you get to ramble and get everything out talk about what you have give it freedom to go look for new ideas or new tools and let it know what good really looks like if you want the goal skill i just used i'm giving it away for free the link is in the description grab it ramble to it about your own project and see what it gives you back once a job like that finally comes out right you never want to work it out from scratch again so you bottle it into a skill so think about a job you do every week the same way every time maybe it's your weekly update turning notes into a post, or checking a page before it goes live.
If you'd explain it to someone new the same way every time, that's a good sign it should be a skill. Claude actually writes the skill for you. It comes with a skill called Skill Creator that's made for exactly this, so it handles the folder, the file, and the wording at the top.
You're not learning to write a skill, you're learning to describe a job well enough that somebody else could do it. There are two more ways to make one, and the second way is better. The first way is to describe the job up front, and you tell Claude, three things.
First, tell it what the job is and when you'd want it done. That when part goes at the top of the skill and it's how Claude knows to use it at the right moment. A lot of people skip that part.
Then the skill works fine, but Claude never uses it because it doesn't know when to. Second, walk it through how you actually do the job in your order. Include the step you always forget and have to go back for.
That step you always forget usually ends up being the most important line in the skill. For me, it's often something really simple. Say I'm trying to grab a transcript from YouTube.
It'll look for all these weird routes, APIs, or third -party tools. There's actually a free tool called YTDLP that will grab transcripts from YouTube and it's pretty reliable.
If we just tell it in the skill to use YTDLP, it goes straight there and works the first time. No wasted time or tokens guessing at other solutions. Third, tell it what the finished thing should look like.
The second way is the one I almost never see people use. Say you've spent an hour with Claude working something out and it finally came out right. Don't close the chat.
Tell it to turn what you just did into a skill so you never have to work it out from scratch again. I like this one better because you already know the process works.
It's all right there in the chat. Claude watched you do it and fix it so you don't have to remember every step yourself. Honestly, most of us are pretty bad at that.
There are two quick things to know. Your skill needs to be a folder with a file called skill .md inside. If it's just a loose file on its own, Claude won't read it.
Also, after you make a new skill, start a new session or restart the app. Then type a slash. and your skill is sitting right there in the list with all the others.
It's your own command, which is pretty cool. To show you how far this goes, I have a content engine skill that's pretty specific. It takes every YouTube video I make and creates a post and an image for LinkedIn.
It needs that post to exist before it can wire up my lead shark automation, which sends my free guide by DM to anyone who responds. Then it also creates a guide on the Cloud Code Club website and it publishes to a couple of other places too. There's really a lot of moving pieces to it, but at a high level, that's what it is.
does. You could easily spend an hour or two on all of that. Now I type one slash command and it runs in the background by itself.
It takes about 10 minutes and I don't have to touch it. The last move is the one that makes every other move keep getting better, and that's refining and iterating. I find that people often don't even know what they're really looking for until they actually see the thing in front of them.
That's why it's about getting a basic version first, then improving it and giving it feedback like you would with a person. What's working and what isn't? What could be better?
What don't you like? What's missing? How would someone actually use this?
And what is the end result you really want? Right now, video editing is really hot. And I see a lot of people trying to build these big video editing tools.
They're saying they want Claude to edit their video for them, add motion graphics and sound effects, and cut out the mistakes. But they're doing it all in one prompt. But Claude doesn't know what a mistake is or what kind of design you want.
So they end up with something low quality or generic. What I tell them is to define what good looks like and give it references of a finished polished product then break it down into smaller steps and work on one piece at a time get the script really good first then get the design right then the storyboard and then work out the animation from the storyboard every time you figure something out along the way update the skill with it next time you run it it already knows i did creative work for big brands for 15 years and good creative work has always been made this way you give a clear brief you look at a first version you give notes and you go again With Claude, every one of those notes can go back into the skill, so you only have to give it once.
Once you've got all of these moves running, what you can do with your time changes. For me, AI has given me choice. I've had agency clients, built SaaS, and sold digital products, and I could do any of them.
Now I get to choose, and I choose to make content and build a community. Earlier, I said that when people ask Claude to edit a whole video in one prompt, they get something generic. So I split my editing into six jobs, saved them in a skill, and Claude edited it.
an entire video without me touching an editing timeline once. I go through all six jobs step by step in this video right here that includes what to tell Claude at each step and what to check. Go watch that one next.
The Hook
The bait, then the rug-pull.
Most people use Claude as a chat box and get a tiny fraction of what it can do. Duncan Rogoff, a former art director for Apple, PlayStation and Nissan, lays out the handful of habits that turn it into a worker that produces output in your voice.
Frameworks
Named ideas worth stealing.
06:06list
The seven-part goal prompt
What you want and why
The quality bar
Tools and go find more
Creative freedom
Check its work three times
Delivery
Restated goal plus work autonomously
A brief structure designed to get out of the model's way on a long autonomous job.
Steal forany long task you want Claude to finish without check-ins
01:26model
Two kinds of instructions
SOP: spell out the steps
End goal: state the destination
Use an SOP when you have a way you like it done, and a clear goal when you don't.
Steal fordeciding how to brief any task
CTA Breakdown
How they asked for the click.
VERBAL ASK
12:18next-video
“I go through all six jobs step by step in this video right here. Go watch that one next.”
Ends on a pointer to a related video about splitting video editing into six jobs saved in a skill.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A former brand video editor splits editing into six separate jobs, saves them as a Claude Code skill, and shows what one finished edit costs in tokens and hours.
A 12-minute tutorial that reverse-engineers a faceless YouTube channel earning $12K/month and rebuilds its entire production pipeline inside Claude Code.
A 15-minute walkthrough of Anthropic's open-source skill that interviews you, writes your agent definition, and deploys a self-improving automation to the cloud — with an honest post-mortem on a $12 first run.