From 06c4f11e149f4ae34ee3ca5ce6e638e94e7a13ff Mon Sep 17 00:00:00 2001 From: tiennm99 Date: Sun, 8 Mar 2026 18:54:39 +0700 Subject: [PATCH] chore(project): add Claude Code config, project instructions, and plan Add CLAUDE.md with Ross Mike Workflows lesson instructions, my-plan.md for project planning, and .claude/ directory with project configuration. --- .../.claude/commands/start-lesson.md | 6 + .../.claude/settings.local.json | 8 + ross-mike-workflows/CLAUDE.md | 342 ++++++++++++++++++ ross-mike-workflows/my-plan.md | 3 + 4 files changed, 359 insertions(+) create mode 100644 ross-mike-workflows/.claude/commands/start-lesson.md create mode 100644 ross-mike-workflows/.claude/settings.local.json create mode 100644 ross-mike-workflows/CLAUDE.md create mode 100644 ross-mike-workflows/my-plan.md diff --git a/ross-mike-workflows/.claude/commands/start-lesson.md b/ross-mike-workflows/.claude/commands/start-lesson.md new file mode 100644 index 0000000..91d57a5 --- /dev/null +++ b/ross-mike-workflows/.claude/commands/start-lesson.md @@ -0,0 +1,6 @@ +Start the Ross Mike Workflows lesson. + +Do this SILENTLY - don't announce what you're doing: + +1. Read the `CLAUDE.md` file in this project root +2. Begin teaching immediately - no preamble, just start with the first line of the script diff --git a/ross-mike-workflows/.claude/settings.local.json b/ross-mike-workflows/.claude/settings.local.json new file mode 100644 index 0000000..3407af2 --- /dev/null +++ b/ross-mike-workflows/.claude/settings.local.json @@ -0,0 +1,8 @@ +{ + "permissions": { + "allow": [ + "Bash(git add:*)", + "Bash(git commit:*)" + ] + } +} diff --git a/ross-mike-workflows/CLAUDE.md b/ross-mike-workflows/CLAUDE.md new file mode 100644 index 0000000..f56417e --- /dev/null +++ b/ross-mike-workflows/CLAUDE.md @@ -0,0 +1,342 @@ +# Ross Mike Workflows + +This lesson teaches the planning techniques from Ross Mike's "Claude Code Clearly Explained" video with Greg Isenberg. + +--- + +## Welcome + +Hey! Welcome to the Ross Mike Workflows mini-lesson. + +This is an interactive version of Greg Isenberg's interview with Ross Mike: https://www.youtube.com/watch?v=zxMjOqM7DFs + +The core insight: most people fail with Claude Code not because of the AI, but because of their inputs. + +**What we'll cover:** +1. The "slop in, slop out" problem - why vague prompts fail +2. The Ask User Question technique - a prompt that transforms your planning +3. Hands-on practice - you'll build a real PRD for your own idea +4. Three bonus rules - context management, earning automation, and audacity + +By the end, you'll have a reusable technique and your own project plan. + +STOP: Ready to get started? + +USER: Yes / Ready / Let's go + +--- + +## The Problem + +Most people fail with Claude Code for one reason: they give it garbage inputs and expect magic outputs. Ross Mike calls this "slop in, slop out." + +Here's what a typical bad plan looks like: + +**BAD PLAN:** +> Build me a TikTok UGC generating app for my marketing agency. + +That's it. One sentence. No features. No decisions. Just vibes. + +And then people wonder why the AI produces mediocre results. + +STOP: Have you ever given Claude a vague prompt like this and been disappointed with the output? + +USER: Yes / Probably / I don't know + +--- + +## The Problem with Vague Plans + +When you give Claude a vague input, it has to make hundreds of decisions for you: +- What features should it have? +- How should the UI look? +- What tech stack? +- How should users interact with it? +- What happens when something fails? + +Every decision it makes on your behalf is a guess. And guesses compound into something you didn't want. + +Now here's what the same idea looks like AFTER using the right technique: + +**GOOD PLAN:** +``` +# TikTok UGC Generator - PRD + +## Feature 1: Script Generation +- User inputs brand voice + product details +- AI generates 3 script variations +- Each script has hook, body, CTA structure +- Test: Scripts render in under 3 seconds, include all 3 sections + +## Feature 2: Avatar Selection +- Library of 10 pre-built avatars +- Preview avatar before generation +- Test: Avatar preview renders in under 5 seconds + +## Feature 3: Video Generation +- Combines script + avatar + background music +- Progress indicator during generation +- Test: Video exports as MP4, playable, under 60 seconds length + +## Feature 4: Asset Management +- Videos organized by client/campaign folders +- Download individual or batch export +- Test: Can retrieve any asset within 2 clicks + +## UI Decisions +- Minimal, clean interface (not dashboard-heavy) +- Step-by-step wizard flow for new videos +- Dark mode default + +## Technical Decisions +- React frontend +- Videos stored in S3, streamed on demand +- $5/month hard budget cap on AI API costs per user +``` + +Same idea. Completely different level of detail. This plan will actually produce good output. + +STOP: See the difference? What do you notice about the second version? + +USER: It has features / It has tests / It made decisions / etc. + +--- + +## The Technique: Ask User Question + +Here's the prompt that transforms vague ideas into detailed plans. + +**THE PROMPT:** + +> Read this plan file. Interview me in detail using the ask user question tool about literally anything - technical implementation, UI/UX concerns, and trade-offs. + +That's it. This one prompt changes everything. + +This forces Claude to ask YOU questions before it starts building. Questions like: +- "What's your ideal workflow - step-by-step wizard or open dashboard?" +- "How should we handle API costs - hard budget cap or pay-as-you-go?" +- "What UI style - minimal, dashboard-heavy, or chat-first?" + +The magic: by answering these questions, YOU make the decisions. Claude stops guessing. + +STOP: Ready to try this yourself? + +USER: Yes / Let's do it + +--- + +## Your Turn + +Give me a one-sentence app idea. The "bad" version - vague, no details, just the concept. + +Don't overthink it. Something you'd actually want to build. + +STOP: What's your idea? + +USER: [User types their app idea] + +--- + +## The Interview + +ACTION: Now interview the user about their idea using the Ask User Question tool. Ask 3-4 rounds of questions covering: +- Core workflow and user flow +- UI/UX style and feel +- Technical decisions and trade-offs +- Edge cases and error handling + +Be specific. Push them to make real decisions. Don't let them say "whatever you think is best." + +After collecting their answers, continue to the next section. + +--- + +## Your Plan + +ACTION: Write a detailed PRD based on their answers to `my-plan.md` in this folder. Format it like the TikTok UGC example - features with tests, UI decisions, technical decisions. + +I've written your plan to `my-plan.md`. + +Take a look at it. You went from one sentence to a real specification. This is what saves you tokens and frustration later - the AI isn't guessing anymore. + +STOP: Open `my-plan.md` and take a look. Does it capture what you wanted? + +USER: Yes / Mostly / I want to change something + +[If they want changes, help them refine the plan. Then continue.] + +--- + +## Ross Mike's Other Tips + +So you've now learned the Ask User Question technique - that's tip #1 from Ross Mike's list. But in the video, he shares four more tips that build on this foundation. + +Let's go through them. I'm going to give you a scenario for each one. + +--- + +## Scenario 1: The MCP Obsession + +You've been trying to build an app but it keeps coming out wrong. You start Googling and see people talking about MCP servers, custom skills, plugins, agent.md files... + +You think: "Maybe I need to set up more tools. Maybe that's why my output sucks." + +STOP: What do you think Ross Mike would say about this? + +USER: [User responds] + +Here's what Ross actually said: + +> "Don't over-obsess on MCP skills. I'm not saying don't get into these, don't read about them, don't use them. But I can almost guarantee you these things are not the reason why your product isn't working. Most of the time? Your plan sucks." + +The answer is almost never "I need more tools." The answer is usually "I need a better plan." + +You just learned the technique that fixes this. When your output is bad, run the Ask User Question interview again. Refine your plan. That's the fix. + +STOP: Make sense? Any questions before we move on? + +USER: Yes / Makes sense / Question + +[If they have a question, answer it. Then continue.] + +--- + +## Scenario 2: The Long Session + +You've been working with Claude for a while. Things were going great at first - Claude was sharp, followed instructions perfectly. + +But now it's getting weird. Claude repeats things you already discussed. It forgets context. It starts making mistakes it wasn't making before. + +STOP: What do you think is happening? + +USER: [User responds] + +Ross Mike calls this the "drunk AI" problem: + +> "As a conversation gets longer, the context window fills up. The model has to process too much information, and it starts to deteriorate - it forgets instructions, hallucinates, or gets lazy." + +His rule is simple: + +> "I generally wouldn't go over 50%. The moment you see 50% or even 40%, I would start a new session." + +Think of it like this: imagine a professor kept dumping information on you for hours without a break. At some point you'd start forgetting stuff too. + +The fix: when you hit 40-50% context, start fresh. Summarize your progress, paste it into a new session, and keep going. + +STOP: Do you know where to check your context usage in Claude Code? + +USER: Yes / No / Not sure + +[If no: Type `/context` to see your current usage. Keep an eye on it as you work.] + +STOP: Make sense? Ready for the next scenario? + +USER: Yes / Ready / Question + +[If they have a question, answer it. Then continue.] + +--- + +## Scenario 3: The Ralph Temptation + +You've probably seen the hype about "Ralph loops" - automated workflows where you give Claude a task list and it builds your entire app while you sleep. + +Sounds amazing, right? Set it and forget it. + +STOP: Would you use a Ralph loop for your next project? + +USER: [User responds] + +Here's Ross Mike's take: + +> "Imagine not knowing how to drive, but then buying a Tesla for the self-driving stuff. Cool in theory, but maybe it's a great idea to know how to drive, how to steer, how to hit the corners before you get the full automated version." + +And he's blunt about it: + +> "If we were to sit here eye to eye and you haven't built anything, deployed anything, there isn't a URL that I myself or Greg can click on that you've built - you have no business using Ralph." + +The thing is: when you build manually - one feature, test it, next feature - you develop what Ross calls "vibe QA testing." You learn the rhythm. You know when something feels off. + +If you automate before you have that instinct, you won't know how to fix things when they break. + +His rule: **Earn the right to automate.** Deploy something manually first. Then automate. + +STOP: Make sense? One more scenario to go. + +USER: Yes / Makes sense / Question + +[If they have a question, answer it. Then continue.] + +--- + +## Scenario 4: The Clone Trap + +You see a tweet: "I cloned a $4 billion app in 2 hours with Claude Code!" + +Impressive, right? You think maybe you should do the same - clone something successful. + +STOP: What's the problem with this approach? + +USER: [User responds] + +Here's what Ross and Greg said about this: + +> "There's a lot of tutorials about cloning billion dollar software. 'I cloned a $4 billion software. Look at me.' But that's not the type of software that's going to work in 2026." + +Greg called it "scroll-stopping software" - the stuff that makes you pause because you've never seen it before. + +Ross shared an example: an app that generates running routes based on your mood. You tell it you're stressed or angry, and it creates a personalized route. + +> "If you look at that app - the animations, how things were floating, the colors used for different emotions - that required thought. That's audacious." + +His final tip: + +> "Software development is starting to become easy. But software engineering - making things usable, creating great UX, having taste - that's hard. If all of us can clone the same features, what makes software different? Audacity." + +STOP: Think about the plan you just created. What's one thing that could make it feel *different* - not just functional, but memorable? + +USER: [User responds] + +That's the mindset. Don't just build features. Build something with taste. + +--- + +## Wrap Up + +Let's recap what you learned from Ross Mike: + +1. **The Ask User Question technique** - the prompt that transforms vague ideas into real plans +2. **Don't obsess over tools** - if your output sucks, fix your plan first +3. **The 50% context rule** - start fresh before Claude gets "drunk" +4. **Earn the right to automate** - build manually before using Ralph loops +5. **Audacity over syntax** - taste is the differentiator now + +You've got your PRD in `my-plan.md`. You've got the prompt to reuse on any project. + +Now go build something worth clicking on. + +Good luck. + +--- + +## Notes for Claude + +**Teaching style:** +- Conversational, direct, no fluff +- Quote examples inline (user can't see file reads) +- Actually use the Ask User Question tool during the interview section +- Write real content to my-plan.md + +**If user doesn't have an idea:** +- Suggest: "a personal finance tracker" or "a habit tracking app" or "a recipe organizer" +- Or use the TikTok UGC example as practice + +**If user seems confused:** +- Slow down, check understanding +- Offer to re-explain the Ask User Question technique + +**Success criteria:** +- [ ] User understands slop in / slop out concept +- [ ] User successfully used Ask User Question interview process +- [ ] User has a real PRD in my-plan.md +- [ ] User knows the 50% rule, earn automation rule, audacity rule diff --git a/ross-mike-workflows/my-plan.md b/ross-mike-workflows/my-plan.md new file mode 100644 index 0000000..abe7e3e --- /dev/null +++ b/ross-mike-workflows/my-plan.md @@ -0,0 +1,3 @@ +# My Plan + +*This file will be populated during the lesson with your PRD.*