content-os（内容操作系统）

Content OS: Multi-Format Content Orchestrator

The "produce everything" button. Give one seed idea → get all content types. Or give long-form content → get it split into short-form pieces.

Quick Start

Forward Mode (Seed → All Content)

User: "Content OS: Statins myth-busting for Indians"

Output:
├── Long-form (quality-passed)
│   ├── YouTube script (Hinglish)
│   ├── Newsletter (B2C - patients)
│   ├── Newsletter (B2B - doctors)
│   ├── Editorial
│   └── Blog post
├── Short-form (accuracy-checked)
│   ├── 5-10 tweets
│   ├── 1 thread
│   └── Carousel content
└── Visual
    ├── Instagram carousel slides
    └── Infographic concepts

Backward Mode (Long-form → Split)

User: "Content OS: [paste your blog/script/newsletter]"

Output:
├── 5-10 tweets (key points)
├── 1 thread (condensed narrative)
├── Carousel slides (visual summary)
└── Snippets (quotable sections)

How It Works

Mode Detection

• Forward Mode: Input is a topic/idea (short text, question, or concept)
• Backward Mode: Input is existing long-form content (>500 words)

Forward Mode Pipeline

PHASE 1: RESEARCH
│
├── PubMed MCP
│   └── Search for relevant papers, trials, guidelines
│
├── knowledge-pipeline (RAG)
│   └── Query AstraDB for ACC/ESC/ADA guidelines, textbooks
│
├── social-media-trends-research (optional)
│   └── Check trending angles, audience questions
│
└── OUTPUT: research-brief.md
    └── Synthesized knowledge with citations

PHASE 2: LONG-FORM CONTENT (Full Quality Pipeline)
│
├── youtube-script-master
│   └── Hinglish script → Quality Review → Final
│
├── cardiology-newsletter-writer
│   └── B2C newsletter → Quality Review → Final
│
├── medical-newsletter-writer
│   └── B2B newsletter → Quality Review → Final
│
├── cardiology-editorial
│   └── Editorial → Quality Review → Final
│
└── cardiology-writer
    └── Blog post → Quality Review → Final

PHASE 3: SHORT-FORM CONTENT (Quick Accuracy Pass)
│
├── x-post-creator-skill
│   └── 5-10 tweets → Accuracy Check → Final
│
├── twitter-longform-medical
│   └── Thread → Accuracy Check → Final
│
└── Extract carousel content from long-form

PHASE 4: VISUAL CONTENT
│
├── carousel-generator
│   └── Generate Instagram slides from key points
│
└── cardiology-visual-system
    └── Infographic concepts (if data-heavy)

PHASE 5: OUTPUT
│
└── Organized folder structure with all content

Backward Mode Pipeline

PHASE 1: ANALYZE
│
└── Parse long-form content
    ├── Extract key points
    ├── Identify data/statistics
    ├── Find quotable sections
    └── Determine topic/theme

PHASE 2: SPLIT (Quick Accuracy Pass)
│
├── Generate tweets (5-10)
│   └── One key point per tweet
│
├── Generate thread
│   └── Condensed narrative
│
├── Extract carousel content
│   └── Key points for slides
│
└── Create snippets
    └── Quotable sections

PHASE 3: VISUAL
│
└── carousel-generator
    └── Generate slides from extracted content

PHASE 4: OUTPUT
│
└── All short-form pieces organized

Quality Gates

Long-Form Quality Pipeline (FULL)

Each long-form piece goes through:

1. scientific-critical-thinking

- Evidence rigor check - Citation verification - Claim accuracy - Statistical interpretation

1. peer-review

- Methodology review - Logical consistency - Completeness check - Counter-argument consideration

1. content-reflection

- Pre-publish QA - Audience appropriateness - Clarity check - Structure review

1. authentic-voice

- Anti-AI pattern removal - Voice consistency - Natural language check

Short-Form Accuracy Pass (QUICK)

Each short-form piece gets:

1. Data Interpretation Check

- Are trial results stated correctly? - Are statistics accurately represented? - Is the study conclusion not misrepresented? - Are effect sizes/NNT/HR correctly stated?

This is a sanity check, not full review. User can iterate manually.

Skills Invoked

Research Skills

Writing Skills

Quality Skills

Visual Skills

Repurposing Skills

Output Structure

/output/content-os/[topic-slug]/
├── research/
│   └── research-brief.md           # Foundation for all content
│
├── long-form/                       # Full quality pipeline
│   ├── youtube-script.md           ✓ Quality passed
│   ├── newsletter-b2c.md           ✓ Quality passed
│   ├── newsletter-b2b.md           ✓ Quality passed
│   ├── editorial.md                ✓ Quality passed
│   └── blog.md                     ✓ Quality passed
│
├── short-form/                      # Quick accuracy pass
│   ├── tweets.md                   ✓ Accuracy checked
│   ├── thread.md                   ✓ Accuracy checked
│   └── snippets.md                 ✓ Accuracy checked
│
├── visual/
│   ├── carousel/
│   │   └── slide-01.png...
│   └── infographic-concepts.md
│
└── summary.md                       # What was produced

Invocation Examples

Forward Mode

"Content OS: GLP-1 agonists cardiovascular benefits"
"Content OS: Statin myths for Indian patients"
"Content OS: When to get a CAC score"
"Content OS: SGLT2 inhibitors in heart failure"

Backward Mode

"Content OS: [paste your 2000-word blog post]"
"Content OS: [paste your YouTube script]"
"Content OS: [paste your newsletter]"

Configuration

What Gets Produced (Forward Mode)

Customization

"Content OS: Statins - only YouTube and tweets"
"Content OS: Heart failure - skip editorial"
"Content OS: CAC scoring - long-form only"

Integration with Existing System

Content OS orchestrates skills that already exist in your system. It doesn't replace them - it coordinates them.

You can still use individual skills directly:

• youtube-script-master for just a script
• x-post-creator-skill for just tweets
• carousel-generator for just slides

Content OS is for when you want everything at once.

Notes

• Long-form content takes longer due to quality pipeline
• Short-form is faster (quick accuracy pass only)
• Research phase runs once, shared by all content
• Visual content generated from text output
• All content uses same research foundation for consistency

Voice & Quality Standards

All content follows:

• YouTube: Peter Attia depth + Hinglish (70% Hindi / 30% English)
• Twitter/Writing: Eric Topol Ground Truths style
• B2B (Doctors): JACC editorial voice
• Anti-AI: No "It's important to note", no excessive hedging
• Citations: Q1 journals, specific statistics, NNT/HR/CI when relevant