15:32Rob Shocks · Tutorial
The Hybrid Claude/Codex Skill Stack That Ships Faster
Rob Shocks audits every major Claude Code skill library and collapses them into one five-step loop you can run from memory.
May 18thThe argument in one line.
In an era when anyone can build anything with AI, the bottleneck has shifted from execution to selection, and a structured validation interview that challenges your assumptions before you write a line of code is now the highest-leverage move a founder can make.
Read if. Skip if.
READ IF YOU ARE…
- You are a solo founder or indie builder with at least one app idea you are about to start building.
- You have used Claude Code, Cursor, or Codex before and are comfortable running slash commands in an agent.
- You have shipped something before but want to avoid building a product nobody pays for.
- You are deciding whether an idea is worth your next 3-6 months before committing to it.
SKIP IF…
- You already have paying customers and are past the validation stage.
- You are building purely to learn -- this framework is designed for market-fit decisions, not learning projects.
The full version, fast.
AI has collapsed the cost of building, which means the bottleneck for founders has shifted from execution to judgment. MoltenOS Validate addresses this by running a five-round interview against your idea using ten criteria distilled from startup frameworks -- problem clarity, pain severity, demand evidence, differentiation, beachhead, business viability, distribution, timing, testability, and founder fit -- then producing a scored report with red flags and specific next experiments. The key insight demonstrated live: a 35/50 score on a real app exposed a missing price model and zero stranger validation in five minutes, giving the founder a concrete list of what to test before wasting months building.
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.
Create a free account →Where the time goes.
00:00 – 00:54
01 · The Focus Argument
AI removed the execution barrier. The new competitive advantage is judgment -- picking the right thing to build. Credits Paul Graham, Eric Ries, Steve Blank as sources.
00:54 – 02:27
02 · The 10 Validation Criteria
Walks through all ten criteria: Problem Clarity, Pain Severity, Demand Evidence, Differentiation, Beachhead/Wedge, Business Viability, Distribution Path, Timing, Testability. Each gets an on-screen title card.
02:27 – 04:00
03 · Skill Setup
Live install of MoltenOS via npx into Codex Desktop. Shows project vs. global scope selection, running /molten validate, switching to high quality mode.
04:00 – 05:14
04 · Sponsor (monday.com)
Sponsored segment demonstrating AI backlog triage agent Mark with full project context and multiple trigger options.
05:14 – 07:31
05 · Roasting His Own Idea (Levercast)
Runs /molten validate against Levercast via voice dictation. Walks through rounds 3-5. Reads the 35/50 verdict live with full scorecard and red flags.
07:31 – 08:06
06 · Kill / Test / Build Decision
Three-outcome framework at end of validation. Revalidation loop after completing next steps. The score gives frame of mind, not a mandate.
08:06 – 09:04
07 · The MoltenOS Skill Chain
Previews the full sequence: molten-brand, molten-design, molten-landing. Closes with Andrej Karpathy recommendation.
Lines worth screenshotting.
- AI eliminated the barrier to building, so the real competitive advantage is now picking the right thing to build -- not building faster.
- A validation score below 25 gets capped regardless of your answers -- the skill refuses to endorse ideas with zero external evidence.
- Founder-market fit can be perfect (5/5) while business viability is 1/5 -- loving the problem does not mean you have the numbers.
- The Mom Test rule applies to AI validation too: your friends, family, and LinkedIn audience are not evidence of demand.
- Voice dictation produces better validation interview answers than typing -- more fluid, more information, less hedging.
- A Wizard of Oz MVP has been replaced by AI-generated prototypes, so you can test assumptions faster than ever at the idea stage.
- Running the validation skill again after completing recommended next steps gives you a second score -- the delta tells you whether you moved the needle.
- An idea with a great distribution path still needs stranger validation before you can trust that people will pay.
- Pricing pulled out of thin air is the fastest way to get a 1/5 on business viability -- unit economics require CAC, willingness to pay, and LTV estimates.
- The beachhead principle requires hyper-specificity: not a CRM for dentists but a CRM for dentists in the DACH region of Germany.
Validate the assumption, not the dream.
WHAT TO LEARN
The riskiest moment in building is before you start -- when your enthusiasm is highest and your evidence is lowest.
- The validation bottleneck has shifted: AI made building trivially easy, so the constraint is now judgment -- picking which idea deserves months of your life.
- Vitamins vs. painkillers is still the fastest first filter -- if someone would only use your product when they happen to remember it exists, the demand is probably not strong enough to build a business on.
- Evidence from strangers counts; evidence from friends, family, and your LinkedIn network does not -- the Mom Test principle holds even when an AI agent is asking the questions.
- A beachhead is not a niche, it is a hyper-niche: not a CRM, not a CRM for dentists, but a CRM for dentists in the DACH region -- the smallest market you can dominate first.
- Business viability requires three numbers -- cost of acquisition, willingness to pay, and lifetime value -- and a price you invented without data gives you a 1 out of 5 on that dimension.
- A score cap below a threshold protects you from confidently answering all the questions and still convincing yourself to build something with zero external validation.
- The revalidation loop is the mechanism: complete the recommended experiments, then re-run the interview -- the delta in score tells you whether you are moving toward or away from product-market fit.
- An AI-generated prototype has replaced the Wizard of Oz MVP for initial assumption testing -- the speed advantage means you can run real experiments earlier than was previously possible.
Terms worth knowing.
- MoltenOS
- An open-source set of AI agent skills for Claude and Codex that walk a founder through validation, branding, design system creation, and landing page generation in sequence.
- Beachhead
- The narrowest possible initial market segment a product should target before expanding -- the principle that you win a small market completely before attacking a larger one.
- Vitamins vs. Painkillers
- Paul Graham framework distinguishing nice-to-have products (vitamins, taken when remembered) from must-have ones (painkillers, taken when in pain). Painkiller demand is more fundable and more defensible.
- The Mom Test
- A customer discovery methodology filtering out false validation -- your mom will say she loves your idea, so you must ask questions that even she cannot lie about, like behavioral evidence of past spending or workarounds.
- Wizard of Oz MVP
- A prototype technique where the user-facing experience looks automated but is run manually behind the scenes -- used to test demand before building real infrastructure.
- CAC
- Cost of Acquisition: the total cost to acquire one paying customer including marketing, sales, and time. Must be significantly lower than LTV for a business to be viable.
- LTV
- Lifetime Value: the total revenue a single customer generates over their entire relationship with the product. Compared against CAC to determine whether a business model is profitable.
- Score cap
- In MoltenOS Validate, a hard ceiling of 25/50 applied to ideas with zero external validation -- prevents founders from getting a false high score by answering confidently with no evidence.
Things they pointed at.
00:25channelPaul Graham / Y Combinator
00:25bookThe Lean Startup (Eric Ries)
00:25channelSteve Blank
01:03bookThe Mom Test (Rob Fitzpatrick)
08:57channelAndrej Karpathy -- what to build in 2026
Lines you could clip.
00:05
“Your biggest advantage is not your AI code factory or your incredible skill set up. It's simply focus.”
00:28
“This skill is not for the faint of heart. It's not going to pat you on the back and say, you're absolutely right or that's a great idea.”
06:12
“A real smack in the face -- an unproven claim that Levercast is meaningfully better than a founder using ChatGPT, Buffer, Typefully, Notion, or manually posting.”
06:48
“Business viability: 1 out of 5 -- purely because I'm suggesting a price here, but it's kind of pulled out of thin air.”
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.
analogystory
00:00In today's market where anyone can build anything, your biggest advantage is not your AI code factory or your incredible skill set up. It's simply focus. Picking the right project to back that's a fit for the market and maybe more importantly, a fit for you as a founder and a builder.
00:18Over the last few years, I've relied and taught frameworks from startup legends like Paul Graham from Y Combinator, Eric Wiese from Startup, Steve Blank, and I've gone and built one skill that encapsulates all of their frameworks into a five minute validation skill you can run-in any agent like codex, clogged code, cursor, whatever you want.
00:36This skill is not for the faint of heart. It's not going to patch you on the back and say, you're absolutely right or that's a great idea. Focus.
00:45It's going to challenge your assumptions and create real experiments for you to follow-up on. And I think it's really gonna surprise you once you run it.
00:53Test number one, problem clarity. Are you completely clear on what problem it is you are solving? Pain severity.
00:59Is this a nice to have or a need to have? Paul Graham calls this the difference between vitamins and painkillers. Number three, demand evidence.
01:06What is the real evidence, not the made up evidence that somebody actually wants this? This is not your friends on LinkedIn, your family, your mom. There's a great book about this called the mom test.
01:16It is actual people willing to either spend money, spend time, or use their reputation to back up your product. Differentiation.
01:24What makes you different from everybody else out there? Your beachhead or your wedge. What specific market are you focused on?
01:30And I mean super niche. So let's say you're focused on a CRM system. It's not a general CRM system.
01:36It's a CRM system for dentists. And not just all dentists, it's for dentists in Germany. And not just all dentists in Germany, it's the DACH region in Germany.
01:45I'm talking about getting super niche down because you need to be able to boil a small market first. It's one thing to have a product that people want, but are they willing to pay for it at a price that drives a profit? What is your cost of acquisition?
01:58What is the lifetime value of the customer? These unit economics really need to add up. How are you gonna find these people and sell to them, and can you do that on a sustainable basis?
02:07Timing. Why is this idea going to win now in a way that it didn't win before? Has there been a regulation change?
02:13Is there a new technology that suddenly allows this? What's the shift in the market?
02:18This is one a lot of people don't think about. What's the easiest way we can test any assumptions we have about our idea or product so that we reduce the risk?
02:29So MoltenOS is a new set of skills in an operating system I'm working on to help you go from idea to market test into product. If you find it useful in any way, hit the star button. You can track it, and any new skills that I'll be adding in will be updated.
02:42So I'm just gonna copy this prompt here. Super simple. And I'm just saying install all these skills in the repo to my project.
02:49Alternatively, you can just pop open the terminal with control j, and I can manually install it by hitting MPX skills add switch dimension MoltenOS core.
02:59It's asking me what skills I want to install. I'm just gonna go ahead and put in validate for now. You can install them all if you wanna check them out.
03:05You can install at the project level or I can install it globally so I can use it on every single project that I want. So once you have the skills installed, just type slash Molten and then you should get all the different versions.
03:15And here we have the Molten validate. If it isn't showing up for you, you need to restart codecs or clog code or cursor just to make sure it shows up. So once I've restarted, I'm just selecting Molten validate here, and I just simply hit return.
03:28Now I'm gonna switch it up to high here to get the best possible response. So to test out the skill, we're going to run it against our app idea Leverkast. You you can capture any idea you have for a post and then use AI to publish it across multiple different platforms.
03:41And this is one of the apps we build out in my community switchdimension.com. So the best way to work with this is to use voice dictation. It's much more fluid.
03:48You're more likely to give more information, then it's going to give you a fast mode or a five minute five round interview mode. I highly recommend the five minutes.
03:56It's just five minutes. It'll get a much better idea of what you're building and give better recommendations. So a quick behind the scenes on how I'm building Leverkast without losing my mind.
04:07So I've got my whole product backlog inside of monday.com, and it's a mixture of features, bugs, and those vague, we should probably do this kind of tasks. Normally, I'd just spend a bunch of time just figuring out what's actually blocking what, but I've been using monday.com's AI agents.
04:23So here I'm saying, build me an agent that helps me triage my product backlog, and I'm giving a reference to the project. So I've got Mark, my backlog triage agent or product manager. It's not just a chatbot.
04:34It's got full context to the project history, conversations, everything that you have within monday. Even your strategy if you want.
04:41So it's making decisions based on the big picture of your company and your product. And you can decide to trigger this agent whenever you want. It can be on a schedule, whenever something changes on the board, or whenever you at mention it in a conversation.
04:52So let's go ahead and run our agent. Hey, Mark. What should we focus on next from our backlog?
04:57So Mark has given me a really helpful steer on what I should work on next across my backlog. So agents are available in monday.com right now.
05:05Check out the link in the description to try it out for free. So you can see here I'm actually roasting my own idea lever cast. So round three of five gets really interesting.
05:15It's asking some of the hard questions. What evidence do you have from strangers, not friends or family or advisers? Have strangers actually described this problem unprompted to you?
05:25Have you interviewed 10 plus people about their current workflows right now, or has anyone actually committed any money, time, or reputation?
05:34So I have to be completely honest. I don't have any preorders for my idea. I don't have any deposits.
05:39I haven't got my pilot launched, letter of intent, anything like that. I need to state that and let the agent know. This isn't a bad thing.
05:46The skill is gonna help us move forward with that as well. So when the process is finished, we get a validation report. So we have our hypothesis and our verdict, our 35 out of 50, which is a pursue cautiously.
05:58Now in the case where you've got no validation whatsoever and no feedback, a score cap will be applied. So you can't actually get above 25. This is actually to protect you that you don't go jumping into an idea that has zero validation.
06:11Basically, it's telling me that this is a good founder fit, but the weakest parts are lack of paid or high commitment demand, which is absolutely true. And a real smack in the face here, an unproven claim that Leverkast is meaningfully better than a founder using ChatGeBT, Buffer, Typefli, Notion, or manually posting.
06:28It's true. I know it's better for me and a couple of other people on the alpha test, but I don't really have any strong meaningful feedback outside of that. So we need to work on that.
06:36Business viability, I'm getting a one out of five purely because I'm suggesting a price here, but it's kind of pulled out of thin air. So I don't know how much it's gonna cost to acquire the customer, so that's the cost of acquisition, how much they're willing to pay, and then how long they're gonna hang around for.
06:51That's the lifetime value. In terms of founder fit, it basically gives me a five out of five. I live the problem.
06:56I run a business in this space. In terms of assumption testability, in the past, I would do a Wizard of Oz style MVP in that I would have some kind of a basic web page.
07:05The user would land on that, and I would do everything manually in the back end. But now because we can generate Apps with AI so much faster, it is easier to test at an earlier stage using some kind of a basic MVP or feature prototype. So finishing up, it pulls together red and yellow flags, things that you really need to think about, your riskiest assumption, and in my case, it's the idea that people will actually pay for this.
07:26How do we test this one assumption before wasting a whole load of time and energy? It give me some next steps in terms of what it recommends that I do in order to move forward. So at the end of this five minute process, you're deciding to kill the idea completely.
07:40You wanna test it a little bit further based on the feedback, or you're ready to go and get stuck in and build it. And once you've gone through the recommended next steps, you just simply run the revalidation again, and it will go and give you your score, whether that's higher or lower than it was before. You don't have to faithfully follow this recommendation or these steps, but it will give you the right mentality, the right frame of mind in order to think about your project or product, whether it's worth pursuing or not.
08:06When you're finished with the validation skill, it is worth checking out the next set of skills. That is Molten Brand, which creates a brand dot m d file. It helps you pin down your positioning, your auto audience, and your messaging.
08:18Then you create a design system. It's targeted at the persona, and then we can go ahead and create a landing page that is designed to be high converting.
08:26It's not one of these standard website landing pages that you get. It's designed to go through a process of storytelling in order to convert your early customers into a call to action, anything from a purchase to a wait list sign up or free trial.
08:39You'll see from the commits, I'm updating and adding new skills on a daily basis based on the feedback I'm getting from my own community at switchdimension.com and from the products that I am building myself. So obviously, if you've made it this far, you're really interested in building apps that people want.
08:53If you want to understand the new meta and what to build in 2026, I highly recommend you check out Andrzej Karpathy's take on what you should be building in 2026. Check that one out next.
The Hook
The bait, then the rug-pull.
AI collapsed the cost of building. Now anyone can ship an app in a weekend -- which means the question is no longer whether you can build, but whether what you build is worth building at all. That reframe is the premise of this nine-minute tutorial, and it hits harder than most founders expect.
Named ideas worth stealing.
00:54model
MoltenOS Validate -- 10-Criteria Scoring System
- Problem Clarity
- Pain Severity
- Demand Evidence
- Differentiation
- Beachhead/Wedge
- Business Viability
- Distribution Path
- Timing / Why Now
- Testability
- Founder-Market Fit
A 50-point scoring system evaluating a startup idea across ten criteria drawn from Paul Graham, Eric Ries, Steve Blank, Alex Hormozi, and Ash Maurya. Ideas with zero external validation are capped at 25.
Steal forPre-build idea evaluation for any product -- run before committing to any significant development investment
08:06model
MoltenOS Skill Chain
- molten-validate (idea scoring)
- molten-brand (positioning + brand.md)
- molten-design (design system)
- molten-landing (high-converting landing page)
A four-step sequence from raw idea to testable landing page, each step a Claude/Codex skill that builds on the output of the previous.
Steal forStructuring the 0-to-MVP workflow for any solo founder building with AI agents
How they asked for the click.
VERBAL ASK
08:06link
“Hit the star button. You can track it, and any new skills that I'll be adding in will be updated.”
GitHub star CTA for MoltenOS repo delivered naturally mid-video during skill setup. Secondary CTA: watch Karpathy video at close.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
