# Scriptwriter

> Write production-ready video ad scripts with full production notes — scene-by-scene breakdowns with dialogue, B-roll, on-screen text, camera direction, hooks, and timing. Outputs individual script files plus a combined approval overview. Use when user says "/scriptwriter", "write a video script", "video ad script", "reels script", "Meta ad script", "scene-by-scene", "écris un script vidéo", "script de pub vidéo", "script reels", "rédige un script", "storyboard de pub".

- Skill: `agentik-os/scriptwriter` (Agent Skill)
- Install (CLI): `npx skillmds@latest add agentik-os/scriptwriter`
- Raw SKILL.md: https://api.skillmd.com/api/skills/agentik-os/scriptwriter/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Marketing & Growth
- Author: agentik-os (https://skillmd.com/u/agentik-os)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/agentik-os/scriptwriter

---


# Scriptwriter

Write production-ready video ad scripts for social/Meta campaigns.

> **Portability note:** Output paths below use `~/clawd/output/{project}-campaign/` (an OmegaOS VPS convention). On any other machine, write to the project's own output dir (e.g. `./output/{project}-campaign/` or wherever the brief specifies) — keep the same internal structure.

## Single-voice craft (do NOT parallelize)

Scriptwriting is **one coherent voice, one creative judgment**. Do NOT fan this out to parallel agents, do NOT run an "audit/verify 2-of-3" loop, and do NOT split scenes across workers — a script's hook, emotional arc, and CTA must be authored by a single mind so rhythm and tone stay continuous. Write all scripts in a campaign sequentially, in the same hand.

**Craft upgrades (apply every time):**
1. **One ear, read aloud.** Every line is performed, not read. Mentally voice each scene at 2.5–3 words/sec; cut any phrase that trips the tongue. Contractions, short clauses, one idea per breath.
2. **Hook owns the first 3 seconds, the arc owns the rest.** Pick ONE hook formula (below) and commit — don't hedge two openings. Then make every scene earn the next: each beat raises a question the following beat answers, so the viewer can't scroll.
3. **Specificity beats adjectives.** Concrete number, named tool, real before/after > "amazing", "powerful", "game-changing". Voice-match the brief's brand; if no brand voice is given, default to plain, confident, conversational — never ad-speak.

## Output Structure

### Individual Script Files (for production)
```
~/clawd/output/{project}-campaign/
├── scripts/
│   ├── 01-{script-name}.md
│   ├── 02-{script-name}.md
│   └── ...
```

### Combined Overview (for approval)
```
├── scripts-overview.md    # All scripts in one doc
```

## Script Template

Each script file follows this structure:

```markdown
# {Brand} Video Ad Script: {Title}

**Duration:** {X} seconds  
**Format:** {Talking head / Screen recording / Animation / Mixed}  
**Aspect Ratio:** {9:16 / 1:1 / 16:9}  
**Funnel Stage:** {TOFU / MOFU / BOFU}  
**Landing Page:** {URL}

---

## 📋 Overview

| Element | Details |
|---------|---------|
| **Hook** | {One-line hook strategy} |
| **Emotion** | {Emotional journey: X → Y → Z} |
| **CTA** | {Call to action} |
| **Tone** | {Voice/energy description} |

---

## 🎬 SCENE 1: HOOK (0:00 - 0:03)

### Shot
- **Type:** {Shot type}
- **Setting:** {Location/background}
- **Energy:** {Mood/energy level}

### Dialogue
> "{Exact words}"

### On-Screen Text
```
{Text that appears on screen}
```

### B-Roll Options
1. {Option 1}
2. {Option 2}

### Direction
- {Camera/performance direction}
- {Timing notes}

---

[Continue for each scene...]

---

## 📝 FULL SCRIPT (Clean Copy)

{Complete script without formatting for teleprompter}

**Word count:** {N} words  
**Reading time:** ~{N} seconds

---

## 🎨 THUMBNAIL CONCEPT

{2-3 thumbnail options with descriptions}

---

## 🎥 PRODUCTION NOTES

{Equipment, recording tips, editing notes}

---

## 📊 VARIATIONS TO TEST

| Element | A | B |
|---------|---|---|
| {Element} | {Version A} | {Version B} |

---

## ✅ PRE-PUBLISH CHECKLIST

- [ ] Audio is clear
- [ ] Captions added
- [ ] End card holds 2+ seconds
- [ ] Landing page works
- [ ] Link in description/comments
```

## Script Structure by Duration

### 15-Second Script (Stories/Reels)
```
[0-2s]  HOOK - Pattern interrupt
[2-8s]  VALUE - One key insight
[8-13s] CTA - Clear next step
[13-15s] END CARD
```

### 25-30 Second Script (Standard)
```
[0-3s]   HOOK - Pattern interrupt
[3-10s]  PROBLEM - Agitate pain point
[10-20s] SOLUTION - Present offer
[20-27s] PROOF + CTA - Social proof + action
[27-30s] END CARD
```

### 45-60 Second Script (Long-form)
```
[0-5s]   HOOK - Strong opening
[5-15s]  PROBLEM - Deep agitation
[15-30s] SOLUTION - Explain transformation
[30-45s] PROOF - Testimonials/results
[45-55s] CTA - Clear call to action
[55-60s] END CARD
```

## Hook Formulas

### Pattern Interrupt
> "If you're still [wrong behavior], you're [consequence]."

### Curiosity Gap
> "I [achieved result] and here's what nobody tells you..."

### Controversial Take
> "Stop [common advice]. Here's what actually works."

### Identity Call-Out
> "This is for [specific person] who [specific situation]..."

### Question Hook
> "Why do [surprising thing] when you could [better alternative]?"

### Result Lead
> "[Company] went from [before] to [after]. Here's how."

## On-Screen Text Guidelines

1. **Keep it short** — 3-5 words max per text element
2. **Reinforce, don't repeat** — Add to dialogue, don't duplicate
3. **Use hierarchy** — Important words bigger/bolder
4. **Time it right** — Appear with dialogue, not before
5. **Position consistently** — Same area throughout video

### Text Styles
```
EMPHASIS: All caps, bold
"Quote style": In quotation marks
Stat/number: Large, accent color
CTA: Arrow or emoji pointing (👇 ➡️)
```

## B-Roll Categories

### Screen Recordings
- Tool interfaces (ChatGPT, Make, etc.)
- Workflow automations running
- Results/dashboards
- Course/product preview

### Talking Head Variations
- Different angles (straight, 3/4)
- Different framings (tight, medium, wide)
- Gestures and reactions

### Stock/Generic
- Person working at laptop
- Team collaboration
- Frustration → success moments
- Abstract tech visuals

### Brand Specific
- Product screenshots
- Customer testimonials (clips)
- Behind-the-scenes
- Event footage

## Emotional Arcs

### TOFU (Awareness)
```
Frustration → Curiosity → Hope
"I'm stuck" → "Wait, what?" → "I could do this"
```

### MOFU (Consideration)
```
Skepticism → Understanding → Confidence
"Does this work?" → "This makes sense" → "I need this"
```

### BOFU (Decision)
```
Hesitation → Trust → Urgency
"Is it worth it?" → "Others got results" → "I should act now"
```

## Pacing Guidelines

| Duration | Words | Pace |
|----------|-------|------|
| 15s | 35-45 | Fast, punchy |
| 30s | 65-80 | Conversational |
| 45s | 100-120 | Room to breathe |
| 60s | 140-160 | Storytelling pace |

**Rule of thumb:** 2.5-3 words per second for natural delivery.

## Workflow

### 0. Optional: analyze reference videos with /watch

When the brief includes reference or competitor videos (URL or local file), run /watch on each one first. The report.md gives you scene-by-scene pacing (cuts per minute, shot lengths), the word-level transcript of the opening, and hook frames. Treat it as the script research: mirror the pacing that holds attention, note where the reference hooks in the first 3 seconds, and steal the structure, not the words. Skip when the brief has no reference videos.

### 1. Receive Brief
From `/campaign_planner` or direct request:
- Ad concept and funnel stage
- Target duration
- Key messaging points
- Landing page
- Brand voice guidelines

### 2. Write Script
- Start with hook (most important)
- Build emotional arc
- Time out each section
- Add B-roll suggestions
- Include on-screen text

### 3. Review & Polish
- Read aloud for timing
- Check word count vs duration
- Ensure CTA is clear
- Verify landing page alignment

### 4. Add Production Notes
- Thumbnail concepts
- A/B test variations
- Equipment suggestions
- Editing notes

### 5. Deliver
- Save individual script file
- Add to overview document
- Send for approval

## File Naming Convention

```
{NN}-{slug}.md

01-free-crash-course.md
02-ai-readiness-quiz.md
03-ai-operator-course.md
04-team-training-case-study.md
```

## Overview Document Structure

```markdown
# {Brand} Campaign Scripts Overview

**Campaign:** {Name}
**Total Scripts:** {N}
**Date:** {YYYY-MM-DD}

---

## Summary

| # | Script | Duration | Funnel | Status |
|---|--------|----------|--------|--------|
| 1 | Free Crash Course | 30s | TOFU | ✅ Ready |
| 2 | AI Readiness Quiz | 25s | TOFU | ✅ Ready |
| 3 | AI Operator Course | 35s | MOFU | ✅ Ready |
| 4 | Team Training | 40s | BOFU | ✅ Ready |

---

## Script 1: Free Crash Course

[Full script content]

---

## Script 2: AI Readiness Quiz

[Full script content]

---

[Continue for all scripts...]
```

## Integration

Works with:
- `/campaign_planner` — Provides creative briefs
- `/creative_director` — Orchestrates full asset creation
- `/ad_designer` — Creates companion image ads
- `/page_designer` — Creates landing pages scripts point to

## Output Contract

Every invocation MUST produce:
1. **One file per script** at `scripts/{NN}-{slug}.md`, each following the Script Template (overview table, scene-by-scene shots/dialogue/on-screen text/B-roll/direction, clean full-script copy, thumbnail, production notes, A/B variations, pre-publish checklist).
2. **One `scripts-overview.md`** with the summary table + every full script inlined, for one-pass approval.
3. A short **delivery note** listing the files written + their absolute paths.

## Verify (before claiming done)

Run this self-check on every script before reporting:
- **Timing:** word count ÷ ~2.75 ≈ stated duration (within ±15%). If off, trim or pad — never mislabel.
- **Structure:** scene timestamps sum to the total duration with no gap/overlap; END CARD present.
- **Files:** every script file exists AND is mirrored in `scripts-overview.md`; naming matches `{NN}-{slug}.md`.
- **CTA + landing page:** present, specific, and the URL matches the brief (do not invent one).
- Read the clean full-script copy aloud once; if any line snags, rewrite it.

## Guardrails (evidence, no hallucination, scope)

- **No invented facts.** Stats, testimonials, results, prices, feature claims, and the landing-page URL come from the brief. If a needed fact is missing, write `{NEEDS: …}` placeholders and flag them in the delivery note — never fabricate a number or quote.
- **Stay in scope.** Write the requested scripts at the requested durations/funnel stages. Don't add extra scripts, change the offer, or redesign the campaign — hand strategy questions back to `/campaign_planner`.
- **No regression.** This upgrade is additive; all original templates, hook formulas, pacing tables, B-roll categories, and emotional arcs remain authoritative.

## Quality Checklist

Before delivering any script:

- [ ] Hook grabs attention in first 3 seconds
- [ ] Emotional arc is clear
- [ ] Dialogue sounds natural (read aloud)
- [ ] Word count matches duration
- [ ] CTA is specific and clear
- [ ] On-screen text reinforces (not duplicates)
- [ ] B-roll suggestions are actionable
- [ ] Thumbnail concepts included
- [ ] A/B variations provided