Youtube Script Master

YouTube Script Master

Unified skill for creating data-driven, evidence-based cardiology YouTube scripts in Hinglish.

This skill CONSUMES data from the research-engine Python pipeline. It does NOT replace that pipeline with manual web searches.

---

CRITICAL: Run Research Pipeline First

Before writing ANY script, the research-engine should have been run to generate:

• Content calendar with prioritized topics
• Demand analysis (what people want)
• Gap analysis (where opportunities are)
• Narrative analysis (what misinformation to address)

cd "/Users/shaileshsingh/cowriting system/research-engine"
python run_pipeline.py --quick    # Quick mode (~10 min)
python run_pipeline.py            # Full mode (~30 min)

---

Complete Architecture

┌─────────────────────────────────────────────────────────────────┐
│           PHASE 1: DATA COLLECTION (Weekly - Python Pipeline)   │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  channel_scraper.py ──► Scrapes 35+ channels (no API needed)   │
│                         Competition, inspiration, belief-seeders│
│                                                                  │
│  comment_scraper.py ──► Downloads comments from top videos      │
│                         Extracts questions and pain points      │
│                                                                  │
│  OUTPUT: /data/scraped/latest_scrape.json                       │
│          /data/scraped/latest_comments.json                     │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│           PHASE 2: ANALYSIS (Python Pipeline)                    │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  demand_signals.py ──► What topics get views/engagement         │
│                        Question themes, demand scoring           │
│                                                                  │
│  narrative_monitor.py ──► Tracks 8 dangerous narratives:        │
│                           1. LDL skepticism                      │
│                           2. Statin fear                         │
│                           3. Insulin primacy                     │
│                           4. Fasting absolutism                  │
│                           5. Supplement superiority              │
│                           6. Seed oil villain                    │
│                           7. Exercise compensation               │
│                           8. Fear mongering                      │
│                                                                  │
│  gap_finder.py ──► Content opportunities                        │
│                    CORRECTION_OPPORTUNITY (misinformation)       │
│                    LANGUAGE_GAP (English→Hindi needed)           │
│                    DEMAND_GAP (questions but no videos)          │
│                    PROVEN_TOPIC (high views in English)          │
│                                                                  │
│  view_predictor.py ──► ML prediction of video performance       │
│                        Ridge regression + TF-IDF on title        │
│                                                                  │
│  OUTPUT: /output/demand_analysis_*.json                          │
│          /output/narrative_analysis_*.json                       │
│          /output/content_gaps_*.json                             │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│           PHASE 3: PLANNING (Python Pipeline)                    │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  idea_combinator.py ──► Seed ideas (300+) × Modifiers (215+)   │
│                         Filters by pillar, archetype, compat    │
│                         Prioritizes by demand + gap scores       │
│                                                                  │
│  calendar_generator.py ──► 100-day content calendar             │
│                            Mon/Wed/Fri schedule                  │
│                            Balanced by pillar and audience       │
│                                                                  │
│  OUTPUT: /output/calendar.json                                   │
│          /output/100-day-calendar.md (Obsidian-ready)           │
│          /output/idea-briefs/*.md (per-video briefs)            │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│           PHASE 4: KNOWLEDGE BUILDING (Per Video)                │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  knowledge_pipeline.py ──► RAG + PubMed in parallel             │
│    ├─► RAG: Your textbooks/guidelines (AstraDB)                 │
│    └─► PubMed: Latest research (NCBI API)                       │
│                                                                  │
│  OUTPUT: Knowledge brief with citations                          │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│           PHASE 5: SCRIPT WRITING (This Skill - Opus)            │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  INPUTS:                                                         │
│  - calendar.json (which topic, why now)                         │
│  - content_gaps.json (opportunity type)                         │
│  - narrative_analysis.json (if debunk: which narrative)         │
│  - knowledge_brief (evidence for claims)                        │
│                                                                  │
│  APPLY:                                                          │
│  - Hinglish rules (70% Hindi / 30% English)                     │
│  - Script structure (hook → body → CTA)                         │
│  - Debunk protocol (if correction opportunity)                  │
│  - 6-point voice check                                          │
│                                                                  │
│  OUTPUT: Complete 15-30 min script in Hinglish                  │
└─────────────────────────────────────────────────────────────────┘

---

Using Research Engine Outputs

Step 1: Check the Content Calendar

# See next 5 topics to create
python calendar_generator.py --show-next 5

# Or read directly
cat /output/calendar.json | head -100

Each calendar entry includes:

• seed_idea - The topic
• modifier - The angle
• gap_score - Why this is an opportunity
• recommended_date - When to publish

Step 2: Check If Debunk Needed

# Get threat ranking of narratives
python analyzer/narrative_monitor.py --threats

# Generate debunk ideas
python analyzer/narrative_monitor.py --debunk

# Get response video ideas for high-reach misinformation
python analyzer/narrative_monitor.py --response

Output includes:

• Which channels are promoting which narratives
• View counts of misinformation videos
• Pre-generated Hinglish hooks for debunk content
• Matched seed ideas for counter-content

Step 3: Check Correction Opportunities

python analyzer/gap_finder.py --corrections

Returns high-reach misinformation videos with:

• Video title and views
• Narratives detected
• Suggested correction format (direct_response, evidence_synthesis, gentle_correction, indian_context)

Step 4: Build Knowledge for Selected Topic

from rag_pipeline.src.knowledge_pipeline import KnowledgePipeline

pipeline = KnowledgePipeline(verbose=True)
brief = pipeline.synthesize_knowledge("Your selected topic")

Step 5: Write Script Using This Skill

With all data ready, apply the rules below.

---

35+ Tracked Channels (Data Source)

The research-engine tracks these channels in target_channels.json:

Competition (Hindi) - Differentiate/Monitor

• Dr Navin Agrawal CARDIO CARE (300K+)
• Cardiac Second Opinion (100K+)
• SAAOL Heart Center (3.4M) - ANTI-PATTERN

Indian Mega Channels - Monitor/Differentiate

• Fit Tuber (7M+)
• Dr Vikas Bangar (1M+)
• Satvic Movement (1M+)
• Dr Biswaroop Roy Chowdhury (4M+) - CRITICAL ANTI-PATTERN

Inspiration (English) - Absorb Techniques

• Peter Attia MD (1.5M+) - PRIMARY MODEL
• York Cardiology (1M+)
• Nutrition Made Simple (1.2M+)
• The Proof with Simon Hill (1M+)
• Dr Ford Brewer (700K+)
• Medlife Crisis (1.5M+)

Belief Seeders - HIGH DEBUNK PRIORITY

• Dr Eric Berg (11M+) - Keto, insulin primacy, statin fear
• Dr Sten Ekberg (3.5M+) - Insulin, fasting
• Dr Ken Berry (2.5M+) - Carnivore, LDL skepticism
• Dr Mark Hyman (3M+) - Functional medicine
• Dr Jason Fung (1M+) - Fasting
• Dr Pradip Jamnadas (1M+) - Popular in Indian diaspora

---

8 Tracked Narratives (For Debunk Content)

The narrative_monitor.py tracks these dangerous beliefs:

When writing debunk content, use the Steelman-Then-Correct Protocol below.

---

Hinglish Language Rules

Word Choice Matrix

Ratio: 70% Hindi / 30% English (technical terms only)

Sentence Patterns

Explanation:

"Cholesterol do type ka hota hai - LDL jo 'bad cholesterol' hai, aur HDL jo 'good cholesterol' hai. LDL zyada ho toh arteries mein jam jaata hai..."

Evidence citation:

"2023 ki ek study, jisme 50,000 Indians the, usme paya gaya ki..."

Practical advice:

"Toh aap kya karein? Simple hai - daily 30 minute walk, dinner 8 baje se pehle, aur sodium kam..."

Transitions (Hindi)

• Point to point: "Ab doosri baat...", "Teen number...", "Sabse zaroori baat..."
• Contrast: "Lekin...", "Haan, magar...", "Yahan twist hai..."
• Emphasis: "Dhyan se suniye...", "Yeh important hai...", "Yeh mat bhooliye..."
• Story: "Ek patient ka case batata hoon...", "Mere saath kya hua..."

---

Script Structure (15-30 min videos)

HOOK (0:00 - 0:30)

Stop the scroll, create curiosity gap.

Patterns:

• Surprising statistic: "80% Indians jo yeh karte hain, unhe heart disease ka risk double hai..."
• Myth challenge: "Aapne suna hoga ki [belief]. Yeh galat hai. Main batata hoon kyun..."
• Story open: "Ek patient aaye mere paas, 42 saal ke. Unka case aapki aankhen khol dega..."
• Direct question: "Kya aap [common thing] karte ho? Yeh aapke dil ke liye kya kar raha hai?"

Rules:

• NO "Namaste dosto" (boring, skippable)
• First 5 seconds = most critical
• Create information gap that MUST be filled

For Debunk Videos, narrative_monitor.py generates Hinglish hooks like:

• "YouTube pe dekha ki LDL kharab nahi hai? Ek cardiologist ki sachai suniye..."
• "Statin se darr lagta hai? Main aapka darr samajhta hoon. Ab evidence dekhte hain..."

INTRO + CREDIBILITY (0:30 - 2:00)

Establish authority, set expectations.

"Main Dr. Shailesh, interventional cardiologist. Pichhle 15 saalon mein hazaaron patients dekhe hain. Aaj main aapko woh bataunga jo main apne patients ko clinic mein batata hoon..."

BODY - Main Content (2:00 - 25:00)

Structure Options:

A. Listicle (3-5 points)

Point 1: [Setup → Evidence → Practical takeaway]
Transition: "Ab doosri baat..."
Point 2: [Setup → Evidence → Practical takeaway]
...

B. Story-driven

Patient case introduction
What happened (tension)
Medical explanation (education)
Resolution
Lessons learned

C. Myth-busting (Debunk Format)

State the myth clearly
Steelman: Why people believe it (from narrative_monitor data)
Evidence: What studies actually show (from knowledge_brief)
Nuance: The complete picture
What to do instead

Engagement Beats (every 3-4 minutes):

• Question to viewer: "Aapko kya lagta hai?"
• Surprising reveal: "Lekin yahan twist hai..."
• Relatable moment: "Aap bhi soch rahe honge..."
• Pattern interrupt: Change pace, tone, or visual cue

SUMMARY + CTA (25:00 - 30:00)

Summary:

• Recap 3 key points (brief)
• One sentence takeaway
• "Agar sirf ek cheez yaad rakhni ho..."

CTA (choose one primary):

• Subscribe: "Is channel pe aisi videos regularly aati hain..."
• Comment: "Apna sawaal neeche likhiye, main jawab dunga..."
• Share: "Kisi apne ko bhejiye jinke kaam aa sake..."

---

Steelman-Then-Correct Protocol (For Debunk Content)

Step 1: Find the Kernel of Truth

Every popular health belief contains something true. Find it.

Step 2: Acknowledge Explicitly

Wrong:

"Yeh log galat hain. LDL clearly causes heart disease."

Right:

"Yeh belief kahan se aayi? Actually, ek valid point hai. LDL alone se poori picture nahi milti. ApoB, particle count, inflammation - sab matter karta hai. Lekin iska matlab yeh nahi ki LDL matter hi nahi karta..."

Step 3: Show the Logical Error

• Oversimplification: "It's not that simple..."
• Cherry-picking studies: "Jab hum ALL studies dekhte hain..."
• Anecdote vs evidence: "Kuch logon ka experience aisa hai, but population level pe..."

Tone: Never Say / Instead Say

---

6-Point Voice Check

Before delivering ANY script, verify all 6:

See voice-check.md for detailed criteria.

---

Evidence Citation Protocol

For Studies

"2023 mein European Heart Journal mein ek meta-analysis aayi - 200 studies, 20 lakh logon pe. Finding? [specific finding]..."

For Guidelines

"ESC guidelines - Europe ke top cardiologists - recommend karte hain ki [specific recommendation]. Kyun? Because evidence shows..."

For Clinical Experience

"Mere practice mein pichhle 15 saal mein, maine [X] cases dekhe hain jahan [observation]..."

---

Quick Reference: Data Files

---

Slash Commands

---

Deprecated Skills

This skill supersedes:

• /.claude/skills/youtube-script-hinglish/skill.md - DEPRECATED
• /.claude/skills/debunk-script-writer/skill.md - DEPRECATED
• /.claude/skills/cardiology-youtube-scriptwriter/SKILL.md - DEPRECATED

Use this unified skill instead.

---

This skill ensures every YouTube script is DATA-DRIVEN (from research-engine) + EVIDENCE-BASED (from RAG+PubMed) + AUTHENTIC (Hinglish voice with 6-point check).

6-Point Voice Check Reference

Purpose

Before delivering ANY YouTube script, verify all 6 criteria. A single failure requires revision.

---

The 6 Checkpoints

1. Authority Check

Question: Would Eric Topol, Peter Attia, or Andrew Huberman say this in Hinglish?

Passing:

• Speaks with earned confidence
• References evidence naturally
• Acknowledges uncertainty where it exists
• Never appeals to mere authority ("Because I'm a doctor...")

Failing:

• Hedges excessively ("Maybe possibly perhaps...")
• Overreaches without evidence
• Uses generic wellness language
• Sounds unsure or defensive

Test: Read aloud. Does it sound like a world-class expert teaching?

---

2. Domain Expert Check

Question: Does this sound like a cardiologist, NOT a wellness guru?

Passing:

• Uses precise medical terminology (explained in Hindi)
• Cites specific studies, guidelines, clinical experience
• Focuses on cardiovascular specifics
• Acknowledges limits of own expertise

Failing:

• Vague "heart health" platitudes
• Generic lifestyle advice without cardiac specificity
• "Toxins" and "inflammation" without specifics
• Claims expertise beyond cardiology

Test: Could a generic health influencer say this? If yes, add cardiac specificity.

---

3. Rigor Check

Question: Would this pass as journal review if translated to English?

Passing:

• Claims are evidence-based
• Citations are specific (year, journal, sample size)
• Acknowledges study limitations
• Distinguishes correlation from causation
• Presents effect sizes, not just p-values

Failing:

• "Studies show..." without specifics
• Cherry-picked evidence
• Oversimplified conclusions
• Ignores counter-evidence
• Presents preliminary data as settled science

Test: Would you cite these sources in a medical journal submission?

---

4. Accessibility Check

Question: Can a 7th grader in Delhi follow this?

Passing:

• Every English medical term is explained in Hindi
• Uses analogies and relatable examples
• Short, clear sentences
• Builds concepts progressively
• Uses concrete examples (patient stories)

Failing:

• Assumes prior medical knowledge
• Jargon without explanation
• Complex sentence structures
• Abstract concepts without grounding
• Information overload

Test: Read to a non-medical family member. Do they follow?

---

5. Non-Preachy Check

Question: Is this explaining, NOT sermonizing?

Passing:

• Presents information, respects viewer choice
• "Yeh data hai, decision aapka"
• Acknowledges lifestyle is personal
• Empowers rather than lectures
• Provides options, not mandates

Failing:

• "You MUST do this"
• Moral judgment on choices
• Guilt-inducing language
• "If you really cared about your health..."
• Condescending tone

Test: Would a patient feel respected or lectured?

---

6. Non-Judgmental Check

Question: Is this evidence-based, NOT lifestyle shaming?

Passing:

• Facts without moral loading
• Acknowledges systemic factors (access, cost, time)
• Focuses on what people CAN do
• Respects economic constraints
• Celebrates small wins

Failing:

• "If you eat this, you deserve..."
• Ignores socioeconomic reality
• Implies laziness or ignorance
• "Just stop eating [food]" as if it's simple
• Blames individuals for systemic issues

Test: Would someone struggling feel motivated or ashamed?

---

Quick Reference Table

---

Revision Protocol

If any check fails:

1. Identify the specific sentences/sections that fail 2. Diagnose which check(s) they violate 3. Revise targeting that specific issue 4. Re-check all 6 points after revision

---

Examples

Failing Example (Multiple violations)

"Aapko cholesterol ke baare mein sochna chahiye. Studies show ki yeh dangerous hai. Aap agar fatty food khate ho toh aapki galti hai. Change your lifestyle!"

Violations:

• Check 2: Too generic ("cholesterol ke baare mein")
• Check 3: No specific studies
• Check 5: "Sochna chahiye" is preachy
• Check 6: "Aapki galti hai" is shaming

Passing Example

"LDL cholesterol - jisko bad cholesterol kehte hain - jab 100 mg/dL se upar jaata hai, toh arteries mein slowly jam hone lagta hai. ESC 2021 guidelines kehti hain ki high-risk patients ke liye target 55 mg/dL hona chahiye. Lekin har insaan ka situation alag hai. Aapka doctor aapki complete picture dekh ke decide karega. Mera kaam hai aapko evidence samjhana - decision aapka hai."

Passes:

• Check 1: Confident, specific
• Check 2: Specific cardiac mechanism
• Check 3: ESC 2021 guidelines cited
• Check 4: Explained in simple Hindi
• Check 5: "Decision aapka hai"
• Check 6: No judgment, acknowledges individual variation

---

Apply this check to EVERY script before delivery. No exceptions.