Cos Courselesson 3

PROGRESS TRACKING

After each step, run inline Python to update ~/.claudepreneur/crash-course/progress.json.

Introduction

Output this EXACTLY:

═══════════════════════════════════════════════════════════════

   ██████╗██╗      █████╗ ██╗   ██╗██████╗ ███████╗
  ██╔════╝██║     ██╔══██╗██║   ██║██╔══██╗██╔════╝
  ██║     ██║     ███████║██║   ██║██║  ██║█████╗
  ██║     ██║     ██╔══██║██║   ██║██║  ██║██╔══╝
  ╚██████╗███████╗██║  ██║╚██████╔╝██████╔╝███████╗
   ╚═════╝╚══════╝╚═╝  ╚═╝ ╚═════╝ ╚═════╝ ╚══════╝
  ██████╗ ██████╗ ███████╗███╗   ██╗███████╗██╗   ██╗██████╗
  ██╔══██╗██╔══██╗██╔════╝████╗  ██║██╔════╝██║   ██║██╔══██╗
  ██████╔╝██████╔╝█████╗  ██╔██╗ ██║█████╗  ██║   ██║██████╔╝
  ██╔═══╝ ██╔══██╗██╔══╝  ██║╚██╗██║██╔══╝  ██║   ██║██╔══██╗
  ██║     ██║  ██║███████╗██║ ╚████║███████╗╚██████╔╝██║  ██║
  ╚═╝     ╚═╝  ╚═╝╚══════╝╚═╝  ╚═══╝╚══════╝ ╚═════╝ ╚═╝  ╚═╝

  LESSON 3: THE ARCHITECTURE UPGRADE

═══════════════════════════════════════════════════════════════

Then say:

Your CLAUDE.md works. Your carousel-copy skill produced real output.

But here's the thing.

As an entrepreneur, you have layers. Your brand voice is different from your offer details. Your offer is different from your ideal client's pain points.

Right now everything is jammed into one file.

Let's fix that.

Then output:

  ┌─────────────────────────────────────────────┐
  │                                             │
  │  LESSON 3: The Architecture Upgrade         │
  │                                             │
  │  ~15 minutes                                │
  │  Goal: Brand files + router + research      │
  │  Win: TWO quality jumps — architecture      │
  │       then real audience research            │
  │                                             │
  │  PROGRESS: ░░░░░░░░░░░░░░░░░░░░ 0/4 steps  │
  │                                             │
  └─────────────────────────────────────────────┘

  STEP 1 — Why multi-file matters

Type 1 to start.

Wait for confirmation.

Step 1: Why Multi-File

Say:

Step 1 — let me explain why we're doing this.

Your CLAUDE.md right now has everything in one place. Name, business, audience, voice, dream. It works. But it's like a resume — surface level.

The Claudepreneur system uses separate files. Each one handles a different concern:

  ┌──────────────────────────────────────────────┐
  │                                              │
  │  YOUR BUSINESS FILES                         │
  │                                              │
  │  brand.md    → How you SOUND                 │
  │               Voice, tone, style, banned     │
  │               language, signature phrases    │
  │                                              │
  │  offer.md   → What you SELL                  │
  │               Your main offer, pricing,      │
  │               transformation, CTA            │
  │                                              │
  │  icp.md     → Your DREAM CLIENT              │
  │               Pains, desires, objections,    │
  │               their exact language           │
  │                                              │
  └──────────────────────────────────────────────┘

Why separate files?

Because a skill that writes carousel copy needs your voice (brand.md) and your audience's pain language (icp.md) — but doesn't need your pricing details (offer.md).

A CTA slide needs your offer (offer.md) — but doesn't need your full brand story.

Separate files = each skill reads exactly what it needs. More detail. Better output.

Then output:

  ╔══════════════════════════════════════════╗
  ║                                          ║
  ║   CONCEPT UNLOCKED                       ║
  ║                                          ║
  ║   Multi-file architecture:               ║
  ║   brand.md = voice                       ║
  ║   offer.md = product                     ║
  ║   icp.md   = audience                    ║
  ║                                          ║
  ║   Each skill reads what it needs.        ║
  ║                                          ║
  ╚══════════════════════════════════════════╝

  PROGRESS: ████░░░░░░░░░░░░░░░░ 1/4 steps

  STEP 2 — Build your 3 brand files

Type next to build them.

Wait for confirmation.

Step 2: Build 3 Brand Files with Quick Interviews

Say:

Step 2 — let's build these. Quick interviews — 2-3 questions each.

Read their CLAUDE.md to pull in anything you can reuse from their existing answers.

Starting with brand.md — how you SOUND.

I already know some of this from your CLAUDE.md. Let me ask a couple more questions to go deeper.

Question 1: How do you want to sound in your content? Casual and direct? Professional and polished? Warm and encouraging? Give me your vibe — and an example of something you'd actually say to a client.

Ask these ONE AT A TIME:

brand.md questions (2-3):

1. "How do you want to sound in your content? Give me your vibe and an example of something you'd actually say."
2. "What words or phrases do you NEVER want to use? What makes you cringe in your niche?"
3. (Optional, only if needed) "Any signature phrases you come back to again and again?"

After brand.md questions, say:

Got it. Let me save brand.md.

Create ./brand-files/brand.md:

# Brand Voice

> How you sound. Read by ALL content skills.

## Voice
[Synthesized from their answers + CLAUDE.md communication style]

## Signature Phrases
[If they mentioned any recurring phrases]

## Language Rules
- **Banned words/phrases:** [From Q2]
- **Tone:** [From Q1]
- **Style:** [Synthesized — sentence length, energy level]

## What I DON'T Sound Like
[From Q2 anti-references, or synthesized from what they avoid]

Then say:

brand.md saved. Now offer.md — what you SELL.

offer.md questions (2-3):

1. "What's your main offer right now? What do people pay you for? Coaching, consulting, a course, a service?"
2. "What transformation does your client get? Where are they before vs after working with you?"
3. "What's the price point? Rough is fine — just so Claude knows we're talking $50 or $5,000."

After offer.md questions:

Create ./brand-files/offer.md:

# Offer

> Your main product/service. Read by CTA and sales-facing skills.

## What I Sell
[From Q1]

## The Transformation
[From Q2 — before/after]

## Price Point
[From Q3 — approximate range]

## CTA
[Synthesized — natural next step for interested people]

Then say:

offer.md saved. Last one — icp.md — your DREAM CLIENT.

icp.md questions (2-3):

1. "Who's your dream client specifically? Not demographics — describe them as a person. What's their day like?"
2. "What keeps them up at night? Their top 2-3 frustrations."
3. "How do THEY describe their problem? Not your words — their exact language."

After icp.md questions:

Create ./brand-files/icp.md:

# Ideal Client Profile

> Your dream customer. Read by ALL audience-facing skills.

## Who They Are
[From Q1 — person-level description]

## Their Pains
[From Q2 — top frustrations]

## Their Language
[From Q3 — exact words and phrases they use]

## What They Want
[Synthesized — desires implied by their pains]

After creating all 3, output:

  ╔══════════════════════════════════════════╗
  ║                                          ║
  ║   ACHIEVEMENT UNLOCKED                   ║
  ║                                          ║
  ║   brand.md: CREATED (voice)              ║
  ║   offer.md: CREATED (product)            ║
  ║   icp.md:   CREATED (audience)           ║
  ║                                          ║
  ║   3 files. 3 layers of context.          ║
  ║   Claude just got way smarter.           ║
  ║                                          ║
  ╚══════════════════════════════════════════╝

  PROGRESS: █████████░░░░░░░░░░░ 2/4 steps

  STEP 3 — The CLAUDE.md router + before/after

Type next to see the big payoff.

STOP HERE. Wait for confirmation.

Step 3: Slim CLAUDE.md to Router + Before/After Demo

Say:

Step 3 — this is the critical teaching moment.

Your CLAUDE.md right now has everything in it. That's fine for a start. But here's the real architecture advice:

Your CLAUDE.md should be SHORT. Under 200 words.

It's a router — it tells Claude WHERE to find the details, not the details themselves.

Think of it like a receptionist. The receptionist doesn't know everything — they know who to connect you to. Your CLAUDE.md says: "For my voice, read brand.md. For my offer, read offer.md. For my audience, read icp.md."

Let me rewrite it.

Read their current CLAUDE.md. Then rewrite it as a slim router:

# CLAUDE.md

## About Me
[2-3 sentences max from their original CLAUDE.md]

## Brand Files
- `./brand-files/brand.md` — My voice, tone, style, banned language
- `./brand-files/offer.md` — My offer, pricing, transformation
- `./brand-files/icp.md` — My ideal client, pains, their language

## Key Rules
- Always read brand.md before writing content
- Match my voice — [1 key characteristic from brand.md]
- Speak to my audience's pains using their language from icp.md

Save the updated CLAUDE.md (overwrite the original).

Then say:

Done. Your CLAUDE.md is now a router — short, clean, points to your 3 brand files.

Now the moment of truth.

Let's run your carousel-copy skill again. But first, I need to upgrade it to read the brand files.

Update .claude/skills/carousel-copy/SKILL.md — change the instructions to:

# Carousel Copy Writer

Read CLAUDE.md first, then read the brand files it points to:
- brand.md for my voice and style
- icp.md for my audience's pain language
- offer.md for my CTA

When I give you a topic:

1. Ask me for a topic if I haven't provided one
2. Write 10 carousel slides:
   - Slide 1: [HOOK] — stop the scroll, call out my audience's pain point using language from icp.md
   - Slides 2-9: [KEY POINTS] — one idea per slide, in my voice from brand.md
   - Slide 10: [CTA] — clear next step from offer.md
3. Keep each slide to 2-3 short sentences max

## Rules
- Read CLAUDE.md + all brand files before writing
- Use audience pain language from icp.md in the hook
- Match my voice from brand.md throughout
- CTA slide references my actual offer from offer.md
- Every slide should stand alone
- No hashtags in the slides

After updating the skill, say:

Skill upgraded. Now run it. Type /carousel-copy and use the SAME topic you used before (or a similar one).

I want you to see the difference.

STOP HERE. HARD GATE. Wait for them to run it and respond.

After they run it and respond:

See that?

The AHA moment. Compare it to what you got before:

• The hook now uses your audience's actual pain language from icp.md — [call out specific example]
• The voice matches brand.md — [call out specific example]
• The CTA references your real offer from offer.md — [call out specific example]

Same skill. Same command. WAY better output.

That's the multi-file architecture. Each file gives Claude more specific context. More context = smarter output.

And your CLAUDE.md? Under 200 words. A router. Not a dump.

But we're not done yet.

Then output:

  ╔══════════════════════════════════════════╗
  ║                                          ║
  ║   ACHIEVEMENT UNLOCKED                   ║
  ║                                          ║
  ║   Architecture upgrade: WORKING          ║
  ║   First quality jump: VISIBLE            ║
  ║                                          ║
  ║   But your icp.md only has what YOU      ║
  ║   know. What about what your audience    ║
  ║   says when you're not in the room?      ║
  ║                                          ║
  ╚══════════════════════════════════════════╝

  PROGRESS: ███████████████░░░░░ 3/4 steps

  STEP 4 — Research agent enriches your icp.md

This next part is going to change what you think AI can do.

Type next to see it.

STOP HERE. Wait for confirmation.

Step 4: Research Agent Enriches icp.md + Second Quality Jump

Say:

Step 4 — here's where it gets real.

Your icp.md right now is good. It has what YOU know about your audience — their pains, their language, their desires.

But here's the thing. You're one person. You know what your clients tell you in conversations. You know what comes up on sales calls.

You DON'T know what they're typing into Reddit at 2 AM. What they're venting about in Facebook groups. What they're asking in forums when nobody's watching.

That's the real language. And it's gold for your content.

I'm about to show you something. You'll see some permission pop-ups — I'm going to search the internet. Live. Right here. Approve them.

Watch this.

Now use WebSearch to research their audience's REAL pain language. Based on their icp.md, run 2-3 targeted searches:

1. Search for Reddit threads, forum posts, or community discussions about their audience's problems. Example searches:

   • "[their audience type] struggling with [pain from icp.md] reddit"
   • "[their niche] biggest frustration forum"
   • "why [their audience] can't [desired outcome from icp.md]"

2. Search for competitor comments or reviews in their space:

   • "[their niche] coach reviews what they wish"
   • "[their audience type] what I tried that didn't work"

3. If relevant, search for their audience's exact language around the transformation:

   • "[their audience] before and after [transformation from offer.md]"

IMPORTANT: These searches MUST be personalized to their specific niche and audience from icp.md. If they're a fitness coach for busy moms, search for busy moms struggling with fitness — not generic "client pain points."

After searching, say:

Did you see that?

I just searched the internet — live, right here in this terminal — and found what your audience is ACTUALLY saying.

Not a Google tab. Not a copy-paste. I went and found it myself.

Here's what I found:

Then present the research findings organized by category:

Pain language — their actual words:

• [Quote or paraphrase 3-5 real pain phrases from the search results — the exact language people use]

Frustrations with existing solutions:

• [2-3 things they've tried that didn't work]

What they ACTUALLY want (in their words):

• [2-3 desire statements using their real language]

Common objections/fears:

• [2-3 things holding them back]

Then say:

See the difference?

Your icp.md said things like "[quote one of their original pain descriptions]."

But your audience is actually saying things like "[quote a raw, specific phrase from the research]."

That's the gap. Your language vs. THEIR language. And when your carousel hook uses THEIR words, it hits different.

Let me upgrade your icp.md with this research.

You'll see a permission pop-up — I'm updating the file you already built.

Read the existing ./brand-files/icp.md and ADD a new section at the bottom. Do NOT replace the original content — add to it:

## Research-Enriched Language (from real audience conversations)

> These phrases come from real online discussions — Reddit, forums, communities.
> Use these in hooks, pain callouts, and audience-facing copy.

### Pain Language (their exact words)
- [3-5 real phrases from research]

### Failed Solutions (what they've tried)
- [2-3 things they've tried that didn't work]

### Desire Language (what they actually want)
- [2-3 desire statements in their words]

### Common Objections
- [2-3 fears or hesitations]

After updating icp.md, say:

Your icp.md just leveled up. It now has YOUR knowledge of your audience PLUS their actual language from real conversations online.

Now let's see what that does to your carousel copy.

Run /carousel-copy one more time. Same topic. I want you to see the difference.

STOP HERE. HARD GATE. Wait for them to run it and respond.

After they run it and respond:

Look at that hook.

[Call out the SPECIFIC difference — the hook should now use raw audience language from the research, not the user's polished version of their pain. Point to exact phrases.]

Version 1 (Lesson 2) — generic, no context.

Version 2 (Step 3) — your voice, your offer, your audience description.

Version 3 (right now) — your voice, your offer, your audience's ACTUAL words.

Three versions. Same skill. Same command. Each one better because of better context.

That's what happened:

• Architecture gave Claude more specific files to read
• Research gave those files real language from real people

And that research agent? That's just one example of what agents can do. An agent went out, searched the internet, found real conversations, and came back with language you can use in your content.

You didn't do the research. Claude did. In about 30 seconds.

Then output:

  ╔══════════════════════════════════════════╗
  ║                                          ║
  ║   LESSON 3 COMPLETE!                     ║
  ║                                          ║
  ║   brand.md  — your voice, captured       ║
  ║   offer.md  — your product, documented   ║
  ║   icp.md    — audience + REAL research   ║
  ║   CLAUDE.md — slimmed to router          ║
  ║   carousel-copy — 3 versions, each       ║
  ║   better than the last                   ║
  ║   Research agent — your first taste      ║
  ║                                          ║
  ╚══════════════════════════════════════════╝

  PROGRESS: ███████████░░░░░░░░░ 3/6 lessons

Gift Unlock

Say:

Gift time.

Brand File Templates — blank templates for each file type. Share them with a friend, use them to start fresh, whatever you want.

Create the gift file at ~/.claudepreneur/crash-course/gifts/brand-templates.md. Read from the course repo's gifts/brand-templates.md and copy it. If the source doesn't exist, create gift content with blank templates for brand.md, offer.md, and icp.md that users can fill in.

Then output:

  ╔═══════════════════════════════════════════════╗
  ║                                               ║
  ║   GIFT UNLOCKED: Brand File Templates         ║
  ║                                               ║
  ║   Blank templates for brand.md, offer.md,     ║
  ║   icp.md. Fill them in for any business.      ║
  ║                                               ║
  ║   ~/.claudepreneur/crash-course/gifts/        ║
  ║                                               ║
  ╚═══════════════════════════════════════════════╝

Wrap Up

Say:

That's Lesson 3. You just:

• Built 3 brand files — voice, offer, audience

• Slimmed your CLAUDE.md to a router (under 200 words!)

• Upgraded your carousel-copy skill to read all brand files

• Saw the FIRST quality jump — architecture makes output better

• Watched a research agent search the internet live and enrich your icp.md

• Saw the SECOND quality jump — real audience language in your hooks

• Got Brand File Templates to share

Two upgrades. Two jumps. Architecture gives Claude structure. Research gives it real-world language.

And that research agent? That's just a preview. In Lesson 5, you'll build your own agent that orchestrates multiple tasks from one command.

Then output:

  ┌─────────────────────────────────────────────┐
  │                                             │
  │  UP NEXT: LESSON 4                          │
  │  Build Skills From Examples                 │
  │                                             │
  │  Your carousel needs a caption.             │
  │  Instead of building one from scratch,      │
  │  you'll paste real captions and Claude      │
  │  builds the skill FROM those examples.      │
  │                                             │
  │  Type /cos-course:lesson-4 to continue              │
  │                                             │
  └─────────────────────────────────────────────┘

Do NOT invoke lesson-4 for them. They type it themselves.

Update progress: lesson 3 complete.

import json, os
progress_dir = os.path.expanduser("~/.claudepreneur/crash-course")
os.makedirs(progress_dir, exist_ok=True)
progress_file = os.path.join(progress_dir, "progress.json")