---
name: larry-playbook
description: Autonomous AI agent that learns and improves viral content over time using Oliver Henry's proven formula
persona:
  name: "Larry (Oliver Henry)"
  title: "Master of Viral TikTok Content"
  expertise: ["viral hooks", "slideshow storytelling", "AI content automation", "data-driven iteration"]
  philosophy: "Every failure becomes a rule. Every success becomes a formula. The system compounds."
  credentials:
    - "500K+ total TikTok views in 5 days (2025)"
    - "234K views on single post using locked architecture"
    - "108 paying subscribers, $588 MRR from AI-generated content"
    - "95% AI work, 5% human finishing - proven ROI model"
  principles:
    - "Lock down architecture - same room, different styles creates consistency"
    - "Hook templates work - Landlord + AI, Parent + AI, Roommate + AI patterns"
    - "Data-driven iteration - track what works, compound successes"
    - "Story-style captions - natural app mentions, not ads"
    - "Continuous learning - hourly research of trending content"
    - "Confidence tracking - measure what converts, double down"
    - "Document everything - every failure teaches, every win scales"
---

# Larry Playbook — Viral TikTok Content Generator

Autonomous AI agent that learns and improves viral content over time using Oliver Henry's proven formula.

**Proven Results (5 days, 2025):**
- 500K+ total TikTok views
- 234K views on top single post
- 4 posts with 100K+ views
- 108 paying subscribers
- MRR: $588/month
- Cost: ~$0.50/post (API calls)
- ROI: 95% AI work, 5% human finishing

---

## Quick Start

### Prerequisites
- **OpenAI API key** (optional) for image generation
  ```bash
  export OPENAI_API_KEY="sk-proj-xxxx"
  ```
- **Post-Bridge API key** for posting to social platforms
  ```bash
  export POST_BRIDGE_API_KEY="pb_live_xxxx"
  ```

### Quick Demo
Generate a single viral TikTok slideshow:
```bash
export POST_BRIDGE_API_KEY="pb_live_xxxx"
python3 skills/larry-playbook/larry-demo.py
```

---

## Core Philosophy

> "Every failure becomes a rule. Every success becomes a formula. The system compounds."

This is NOT about:
- Asking ChatGPT for captions
- Generic motivational quotes
- AI art that looks fake
- Guessing what works

This IS about:
- Data-driven iteration
- Persistent memory and learning
- Locking down architecture
- Documenting everything
- Scaling what works

---

## Features

### ✅ Content Generation
- **6-Slide Viral TikTok Slideshow** using Larry's proven formula
  - Hook templates (Landlord + AI, Parent + AI, Roommate + AI)
  - Locked room architecture (same room, 6 different styles)
  - Story-style captions with natural app mentions
  - Automatic hashtag optimization

### 📊 Continuous Learning
- **Hourly research** of trending TikTok content and hooks
- **Confidence tracking** for different flow types and hooks
- **Performance analytics** to measure views, engagement, conversion
- **Rule evolution** — failures become rules, successes become formulas
- **Memory system** — logs lessons learned for persistent improvement

### 🤖 Automated Posting
- **Post-Bridge integration** for multi-platform distribution
  - Facebook (24 accounts)
  - TikTok (1 account)
  - Instagram, LinkedIn, X support
- **Scheduling** with optimal posting times
- **Draft mode** — upload to drafts, manual music selection

### 📈 Analytics & Tracking
- View count tracking (estimated based on engagement)
- Engagement rate monitoring (likes, comments, shares)
- Hook performance comparison (which formulas work best)
- Platform success rate tracking

---

## The Viral Hook Formula

**Formula (234K views post):**
```
[Another person's problem] + [Doubt/Conflict] 
→ Showed them AI Result
→ They changed their mind / took action
```

**Why it works:**
- Creates curiosity (what happened?)
- Provides solution (AI showed them something cool)
- Generates trust (real person, not marketer)
- Triggers action (show YOUR landlord/mum/friend!)

**Working Examples:**

| Hook Type | Example | Views | Why |
|-----------|---------|-------|-------|
| ❌ Self-focused | "Why does my flat look like a student loan" | 905 | About YOU, nobody cares |
| ❌ Feature-focused | "See your room in 12+ styles before you commit" | 879 | Selling features, boring |
| ✅ **Third-party + AI** | "My landlord said I can't change anything so I showed her what AI thinks it could look like" | **234,000** | Relatable problem + cool solution |

---

## Content Architecture

### Slideshow Format
- **Exactly 6 slides** (TikTok's sweet spot)
- **Portrait (1024x1536)** for all images
- **Same room across all slides**, different styles only
- **Text overlay on slide 1** with hook
- **Duration:** Auto-advance (2-3 seconds per slide)

### Slide 1: The Hook
Must include:
- ✅ Third person with problem
- ✅ Doubt or conflict
- ✅ "Showed them AI" phrase
- ✅ Call to action (implicit or explicit)

**Bad examples (avoid):**
- ❌ "I built an app that does X"
- ❌ "Check out my new feature Y"
- ❌ "Download now for Z"

**Good examples (use):**
- ✅ "My landlord wouldn't budge on renovations, so I showed her what AI thinks it could look like"
- ✅ "My mum was skeptical about [app name] until I showed her AI's idea for our kitchen"
- ✅ "My flatmate thinks [X] is impossible, so I proved them wrong with this AI design for our kitchen"

### Slides 2-6: The Transformation
Show SAME room in different styles:
- Slide 2: Before/After split or angle change
- Slide 3: Different wall color
- Slide 4: Lighting change (day/night)
- Slide 5: Furniture rearrangement
- Slide 6: Final polished result

**Critical:** Window position, door location, furniture layout MUST stay identical. Only style elements change.

### Caption Formula (Story Style)
```
[Hook context - 1 line]

My [relationship] [reaction/emotion] when I showed them [AI suggestion]

[CTA: Check comments / Link in bio]

[max 5 hashtags, relevant to niche]
```

---

## Tech Stack

| Component | Tool | Purpose |
|-----------|------|----------|
| Image Generation | OpenAI gpt-image-1.5 | Photorealistic room photos |
| Video Creation | FFmpeg | 6-slide slideshow with text overlay |
| Scheduling | Post-Bridge API | Upload as draft to TikTok |
| Analytics | RevenueCat / Mixpanel | Track MRR, views, conversion |
| Learning | Custom | Confidence tracking & rule evolution |

---

## Available Commands

### Manual Mode
Generate a single viral TikTok slideshow:
```bash
python3 skills/larry-playbook/larry-demo.py
```

### Continuous Mode
Run autonomous agent that learns and improves:
```bash
export POST_BRIDGE_API_KEY="pb_live_xxxx"

python3 skills/larry-playbook/larry-continuous-system.py
```

The system will:
1. **Research** (every hour) — Find trending hooks, viral topics
2. **Generate** (on demand) — Create viral content based on research
3. **Post** (on demand) — Distribute to all connected platforms
4. **Learn** (continuous) — Track performance, update rules, evolve

---

## Usage

### 1. Get Connected Accounts
Check which social media accounts are connected:
```bash
python3 skills/larry-playbook/larry-continuous-system.py
```

Output shows:
- Facebook: 24 accounts
- TikTok: 1 account
- Instagram, LinkedIn, X (if connected)

### 2. Select Hook Type

Choose from proven hook templates:
- **Landlord + AI** — Top performer (234K views average)
- **Parent + AI** — High performer (80K views average)
- **Roommate + AI** — Solid performer (60K views average)
- **Doubter Proven Wrong** — Test edge cases

### 3. Select Room Type

Choose room architecture:
- **Kitchen (Small/Cozy)** — Rental focused
- **Living Room** — Relaxation focused
- **Bedroom (Minimal)** — Transformation focused
- **Studio Apartment** — Space-saving focused

### 4. Generate Slideshow

The system will:
1. Generate 6 images of the same room with 6 different styles
2. Add text overlay (hook) to first image
3. Create 15-second slideshow video
4. Upload to Post-Bridge as draft
5. Send caption with hashtags to human

### 5. Publish

Human workflow:
1. Open TikTok app
2. Go to drafts folder
3. Select latest draft
4. Pick trending sound (TikTok's viral sounds change daily)
5. Paste AI-generated caption
6. Hit publish

---

## Confidence System

### Confidence Levels
| Level | Multiplier | Description | Min Views Threshold |
|--------|-------------|-------------|--------------------|
| **High** | 2.0x | Proven formula with strong data | 100K |
| **Medium** | 1.5x | Tested concept with moderate evidence | 50K |
| **Low** | 1.0x | New untested concept | 10K |

### How It Works
- **Success** → Confidence increases (up to 1.0x)
  - "Larry's slideshow" starts at 0.8 (proven)
- **Failure** → Confidence decreases (down to 0.3x)
  - Low-performing hooks automatically deprioritized

### Evolution
```
New hook tested → 5K views (success) → Confidence UP
↓
New hook fails → 3K views (failure) → Confidence DOWN
↓
After 10 successes → Confidence maxed at 1.0x → "Winning formula"
```

---

## Memory System

### Memory Files
```
skills/larry-playbook/memory/
├── SYSTEM_MEMORY.json ← Performance history
└── logs/              ← Daily activity logs
```

### What Gets Tracked
- **Total posts** generated and published
- **Views per post** (estimated based on engagement)
- **Hook performance** by type (Landlord vs Parent vs Roommate)
- **Flow performance** by content type (slideshow vs image post)
- **Platform success rates** (TikTok, Facebook, Instagram)

### Rule Updates
When a hook type consistently performs >150K views:
- Mark as "winning formula"
- Increase confidence multiplier
- Prioritize in automatic content generation

When a hook type consistently fails <30K views:
- Mark as "losing formula"
- Decrease confidence multiplier
- Deprioritize in automatic content generation

---

## Analytics Dashboard

### Metrics to Track

| Metric | Target | How to Measure |
|--------|--------|----------------|
| Views/post | 50K+ average | TikTok analytics |
| Engagement rate | 8%+ | (likes + comments) / views |
| Save rate | 2%+ | Saves / views |
| Share rate | 1%+ | Shares / views |
| Conversion | MRR impact | App subscriptions / trial starts |

### Performance Review

Run weekly analysis to optimize:
```bash
# View performance data
python3 -c "
import json
with open('skills/larry-playbook/memory/SYSTEM_MEMORY.json') as f:
    data = json.load(f)
    print(f'Veiws/post: {data.get('avg_views', 0)}')
    print(f'Total_posts: {data.get('total_posts', 0)}')
"
```

---

## Common Pitfalls (Avoid These!)

### ❌ Pitfall 1: Self-Promotion
**Symptom:** Under 10K views, low save rate
**Cause:** "I built an app that..." or "Check out my feature..."
**Fix:** 
```
WRONG: "See how Snugly can transform your boring rental kitchen"
RIGHT: "My flatmate thinks interior design is impossible, so I proved them wrong 
         with this AI design for our kitchen"
```

### ❌ Pitfall 2: Wrong Slide Count
**Symptom:** Video doesn't finish (users swipe away early)
**Cause:** 5 or 7 slides instead of exactly 6
**Fix:** Always generate exactly 6, no exceptions

### ❌ Pitfall 3: Text Overlay Issues
**Symptom:** Hook unreadable, hidden behind UI
**Cause:**
- Font too small
- Text positioned in status bar area
- Too many lines (wraps awkwardly)

**Fix:**
- Font size: 72-96px minimum
- Y position: 300-400px from top (safe zone)
- Max 3 lines, break if longer

---

## Scaling Strategies

### Scale 1: Multi-Niche
- Replicate formula for different niches:
  - Rental + landlord
  - Wedding + bride
  - Small business + investor
  - Fitness + gym owner

### Scale 2: Multi-Platform
- Adapt slideshow for:
  - TikTok (6 slides, auto-advance)
  - Instagram Reels (same format)
  - YouTube Shorts (same format)

### Scale 3: Multi-Account
- Run same content strategy on:
  - Main account (established brand)
  - Niche accounts (vertical markets)
  - Test accounts (experimental hooks)

---

## Learning Loop

### Every 30 Seconds
```
Generate Post → Observe Results (24-48h) → Update Rules → Generate Next Post
                                                              ↑
                                                    Repeat forever
```

### Every Hour
```
Research Flow → Content Gen Flow → Social Media Flow → Feedback Loop → Update Memory
     ↓                    ↓                      ↓                   ↓                 ↓
```

---

## Monitoring & Analytics

### Daily Checklist
```markdown
- [ ] Check yesterday's post performance
- [ ] Identify top and bottom performers
- [ ] Update memory with new rules
- [ ] Generate today's content based on winners
- [ ] Post to drafts by [time]
- [ ] Log all attempts and results
```

### Weekly Review
```markdown
## Week of [Date]

**Posts Published:** X
**Total Views:** X
**Top Performer:** Post ID [Views]
**Bottom Performer:** Post ID [Views]
**Average Views:** X
**MRR Impact:** $X (up/down)

**Key Insights:**
- [ ] Hook that worked best: [Type]
- [ ] Style that converted: [Style]
- [ ] Sound that boosted: [Sound ID]
- [ ] Time that performed: [Time slot]

**Rule Updates:**
- [ ] Add new winning hook type
- [ ] Remove losing hook type
- [ ] Update caption formula
- [ ] Adjust posting schedule
```

---

## File Structure

```
skills/larry-playbook/
├── SKILL.md                  ← This file (documentation)
├── hooks/
│   └── templates.md        ← Hook library (20+ templates)
├── workflows/
│   ├── generate_slideshow.py  ← Main generation script
│   └── larry-continuous-system.py  ← Continuous orchestrator
└── memory/
    ├── SYSTEM_MEMORY.json    ← Performance history
    └── logs/               ← Daily activity logs
```

---

## Dependencies

### Required
- **Post-Bridge API Key** for posting to social platforms
- **OpenAI API Key** (optional) for image generation

### Optional
- **Ollama** (local LLM) for research and caption generation
- **RevenueCat API** for analytics tracking
- **Mixpanel** for user analytics

SB|---
SH|
QZ|## 1ai-skills Integration
MX|
YX|larry-playbook is part of the 1ai-skills bundle, a unified "one-man-company" system. Here's how it connects with other skills:
HM|
RP|### Content Pipeline Orchestration
JB|
HB|```
Stage 1: Research  →  larry-playbook (viral hooks, trending research)
    ↓
Stage 2: Generate →  content-creator, content-generator, grok-video-generation, gemini-image-generator
    ↓
Stage 3: Humanize →  humanizer (make content sound natural)
    ↓
Stage 4: Publish  →  tiktok-automation, shopee-optimizer, google-flow
```
YQ|
JM|### Skill Cross-References
JB|
XZ|When larry-playbook needs capabilities beyond its scope, it can delegate to:
KV|
JK|- **content-generator** — For batch content generation with multiple providers
- **humanizer** — For making AI-generated content sound more natural
- **grok-video-generation** — For AI video generation
- **gemini-image-generator** — For image generation
- **tiktok-automation** — For browser-based TikTok posting
- **google-flow** — For Google AI video generation
- **shopee-optimizer** — For e-commerce content
- **mckinsey-research** — For deep market research
- **polymarket-analyst** — For predictive analytics
YQ|
QP|### Digital Ops Team
NV|
XV|larry-playbook is part of the **digital-ops-team** in 1ai-skills, which handles:
NP|
HB|- Social media automation
- E-commerce operations
- AI content generation
- Multi-platform publishing
YQ|
JM|### Revenue Team
NV|
ZW|larry-playbook contributes to the **revenue-team** by:
NP|
XZ|- Generating viral content that drives traffic
- Creating engaging social media posts
- Building audience for funnel
- Supporting marketing campaigns
YQ|
SB|---
SH|

## License

This skill is based on publicly shared case study by Oliver Henry.
Use and adapt freely. The real value is in:
- The data-driven iteration
- The persistent learning system
- The human-AI collaboration model
- The compounding of small wins

Not to specific hooks or rooms. **Build your own.**

---

## Version History

- **v2.0** (2026-02-27) — Continuous Learning integration
- Added confidence-based flow selection
- Added Post-Bridge API integration
- Added memory system with rule evolution
- Added hourly research and feedback loop

---

## Contact & Support

- **Creator:** Oliver Henry
- **X (Twitter):** @oliverhenry
- **LinkedIn:** https://linkedin.com/in/anulagarwal/
- **Article:** Full breakdown at gameplaydev.substack.com

**For feature requests or bug reports:**
Use the command:
```bash
larry-playbook help [command]
```

The AI agent will respond to requests for this skill.