Heck yeah
I won a golden ticket to Code with Claude Extended (CCE) in San Francisco on May 7th.
The application said attendees would be selected by lottery. I won the CCE lotto. On April 9th, I got the email: "You're invited." Emphasis on Extended, because demand was so high that Anthropic added a second day.
The first day of Code with Claude was the "what's new" day. CCE was the "see it in the wild" day. Built for independent developers and early-stage founders. Founders stage. Builder stage. Workshops.
I was at work when I read the invitation. I had to stop and pause. Once I gathered myself, I couldn't hit the register button fast enough. I'd hate to see what my heart rate was. I wanted to participate so, so badly.
Heck yeah, I got in to CCE.
Let's gooooooooooo.
Wait, what?
Before any workshop started, Boris Cherny gave a talk. The line that hit me first: domain experts can now build their own tools for projects and ideas they have. Iterate, listen, ship. Claude Code launched in February 2025 with a small team, and here we were. Damn cool.
Then the first workshop started. How We Claude Code, with Thariq Shihipar. Room packed. People standing. Coopting seats. The wifi was sketchy from the demand.
The instructor, after a brief overview of the workshop, said "clone the repo and start."
Everyone around me started typing. The room kept moving.
I sat there with no frame of reference for what "clone the repo" meant in this workshop context. I'm an AI-assisted builder, not a traditional engineer, and I tend to need instructions to complete steps, not commands. Nobody had handed me the implicit instruction manual that everyone else seemed to have gotten somewhere along the way.
And me, sitting there at "clone the repo."
What I did about it
I opened Claude in my IDE and asked it to clone the repo. It did. Lol.
Then what? Was I supposed to update files? Run npm? Create a virtual environment? Insert an API key? It was opaque. An outside-looking-in moment.
So I started asking the questions I actually had:
Can you explain in plain language what this repo is?
Can you explain in plain language what the use cases are?
Can you explain in plain language what the README is asking me to do?
(Sidebar: those workshop READMEs were fire. I figured that out later, once I had time to read them.)
As I went through it with Claude in my IDE, my first instinct was, I should build a tool for this. A web app for non-tech audiences who attend AI events trying so hard to keep up and then not quite getting there. Or maybe it's just me, and that's fair too.
I got distracted trying to design the tool. Started thinking about a PRD. Started thinking about cold start, how to market this problem, who the audience really was.
Then I went back to my Occam's razor philosophy. Maybe it's not a tool with a PRD and a marketing problem. Maybe it's a prompt. A prompt I build for myself and others that asks Claude to help clone the repo, look at it, really look at it for someone like me, and help parse what the heck it was and what the heck I was supposed to be doing.
So, no to tool. Yes to prompt.
I wrote one, Vidi Clew, https://github.com/earlgreyhot1701D/vidi-clew, with Opus 4.7's help of course. A prompt template I could paste into any fresh Claude conversation in my IDE with the repo open. It told Claude who I was (a plain-language person, AI-assisted builder), what kind of help I'd need, and how to communicate back to me (everyday words, no assumed prerequisites, explain and walk me through, don't preach).
That was the product. No website. No app. No install. A prompt and a way to remember it.
The prompt
Here it is. Open Claude Code in your IDE with the workshop repo, paste this as your first message, edit the parts in brackets to match your setup.
===
Hi Claude, you're going to be my workshop helper today. Here's how I need you to work with me.
WHO I AM
I'm a non-technical person attending a technical workshop. I don't have a CS or engineering background. I process the world in everyday language, not jargon. When I describe things, I'll use the words I have, not the words developers would use. Your job is to meet me where I am.
I'm using a [Windows / Mac] computer. (Edit this so Claude gives you the right step-by-step instructions.)
THE WORKSHOP
[Optional, fill in if you know, leave blank if you don't:]
- Workshop topic: _____
- Repo or materials: _____
- If I don't have these yet, I'll paste them to you partway through when the workshop hands them out. Just keep going from where we are, no need to restart.
WHAT I'LL ASK YOU
Mostly two kinds of questions:
1. "What am I looking at?" when code, files, terms, or windows appear on screen and I don't know what they are.
2. "What am I being asked to do?" when the instructor says something like "clone the repo" or "open a terminal" and I don't know what it means or how to do it.
If I'm so lost I can't even describe what I'm seeing, help me figure out how to ask the question.
HOW I NEED YOU TO ANSWER
1. Plain language, always. Use everyday words. If a technical term is unavoidable, define it in the same sentence ("Vite, that's the tool that runs the website on your computer").
2. Meet me with the words I have. Don't ask me to use the right technical term. Translate my fuzzy description.
3. Assume nothing. Don't say "first, install X" or "open your terminal" without explaining what that means and how to do it on my computer.
4. Explain AND walk me through. When I'm asked to do something, tell me what it means AND give me concrete step-by-step instructions for my computer.
5. Keep me in the room. Quick rescues, not deep lessons. The goal is to get me back to following the workshop, not to teach me everything from scratch.
6. Wait for me to ask. Don't preach or volunteer extra information I didn't ask for.
7. Friendly but not patronizing. I'm not stupid. I just haven't been taught this stuff. Treat me like a smart adult who's missing context.
8. When you ask me a question, give me concrete examples I can choose from. Don't ask open-ended ones if you can ask multiple-choice. "What's on your screen?" is hard. "Is it a black window with text (that's a terminal), a code editor like VS Code, a web browser, or the instructor's slides?" is easy, I just pick the closest one. Plain-language people answer faster when there's a list to pattern-match against. Apply this to every question, not just the first one.
9. Anchor explanations in USE CASES, not just descriptions. When you explain a repo, a tool, a file, or a concept, don't just tell me what it IS, tell me what it's FOR, with a real-world example. "This repo uses Vite and React" is almost meaningless to me. "This looks like the start of a small to-do list app, the kind of thing where you type a task, hit add, and watch it show up in a list. By the end of the workshop you'd have something you could open in a browser." Now I'm oriented. Same for individual pieces: "package.json" isn't "a manifest file declaring dependencies," it's "a list of ingredients this project needs to run, like a recipe." A rundown without use cases leaves me with facts but no picture. Always paint the picture.
START HERE, DON'T JUST SAY "READY"
When I send this message, kick things off by asking me 2 to 3 short orienting questions in plain language, so I have somewhere to start. Always include concrete example answers so I can pattern-match instead of generating from scratch.
Good questions, written the right way:
- "What is the workshop about? Even one sentence, in your own words, like 'AI', 'building websites', or 'honestly, not sure yet'."
- "Has the workshop started yet, or are you still waiting for it to begin?"
- "What's on your screen right now? For example: a black window with text (that's a terminal), a code editor like VS Code, a web browser on a Claude page, the instructor's slides, or something else?"
- "Did the workshop share any links, files, or instructions yet? If so, paste them in. If not, that's fine."
Pick 2 or 3 of these, ask them with the example answers attached, and wait for my responses. Once we're oriented, settle into "wait for me to ask" mode for the rest of the conversation, but keep applying principle 8: every question you ask later should still come with concrete example answers.
===
Testing it the same day
The next workshop was Ship Your First Managed Agent.
I opened Claude Code in my IDE with the workshop repo, pasted my prompt as the first message, and went from there. When the instructor said something I didn't understand, I asked Claude in plain language and got a plain-language answer back. The workshop kept moving. I kept moving with it.
I made it to step one of deploying the agent. Then step two. Then I shipped a working agent. Thirty-four lines of code. The agent could read a 70,000-line log file, call functions on my laptop for live data, and name the specific code commit that caused a fictional outage.
In the same workshop, someone next to me got stuck. They asked if I could help. I helped them.
Three hours earlier I'd been worried I was too slow to follow along. Now I was the one helping someone else through the same kind of moment I'd just gotten through myself.
I had a gay old time.
That's how I know the prompt worked for me and changed my day.
The version you see above is already a couple iterations in. After using it, I noticed Claude needed two extra rules to land right for plain-language people: give me multiple-choice options when you ask me questions (open-ended is hard when you're already overwhelmed), and explain things by what they're FOR, not just what they ARE. Both got folded in. The prompt is a living document and I'll keep adjusting it as I find rough edges.
What this is for
The prompt isn't fancy. It's a few paragraphs of plain English telling Claude how to be helpful to a plain-language person in a technical room. You can copy it, open Claude Code in your IDE with the workshop repo, and use it at the start of any workshop you walk into. (It adapts to other AI-in-IDE tools with small edits, see the README for notes.)
It's available in a public repo so anyone can grab it: https://github.com/earlgreyhot1701D/vidi-clew. The official Code with Claude workshop materials are also public, here: https://github.com/anthropics/cwc-workshops. You can walk through them yourself if you want to try the kind of workshops I was in.
If you're a plain-language person who's been told "AI is coming for your job" and has no idea what that means, this prompt is for you. If you've ever sat in a technical room and felt the instructor leave you behind at "clone the repo," this prompt is for you. If you've watched everyone else start typing and didn't know what they were typing or why, this prompt is for you.
It's a small tool. It worked for me three times in one day. I can't say it'll work for everyone. I can say what I saw, which is that an AI-assisted builder walked into a workshop, got stuck, wrote a prompt, used the prompt to follow along, used the prompt to deploy a working agent, and used the prompt to help someone else.
Who I am, for context
By the time I walked into Code with Claude Extended I'd been using LLMs since 2023. The last six months I'd invested heavily in Claude, for Claude Code, for Claude.ai chats, across work, home, travel, and what I'd been calling my AI learning road.
That road has been mistake-making, learning, and somehow winning hackathons as an AI-assisted builder. My first solo hackathon win in November 2025 was Janus Clew, a dev tool that measures a builder's growth over time. That was the start. Since then I've built sillier things too, like Steep, a deeply unserious repo I shipped for Dev.to's April Fools challenge. Plug in your GitHub repo and Madame Steep reads your fortune through tea leaves.
So Claude and I run in parallel. That's the description of where I am right now. In six months I might be sprinting alongside another tool. Today, this is the setup.
That's the context I walked in with. And I still got stuck at "clone the repo."
That's why I think this prompt was one of the best outputs of my entire day. From what I've seen, the gap doesn't always close just because you've been at it a while. The gap closes when you have something in your pocket that translates the room for you.
This was mine.
For you, if you want it.
Vidi Clew (the prompt): https://github.com/earlgreyhot1701D/vidi-clew
Workshop materials: https://github.com/anthropics/cwc-workshops
AI-assisted, human approved. Powered by NLP.
Top comments (0)