Short-Form Script

ok script,' 'short form,' 'video script,' 'hook ideas,' 'script for reels,' 'film this.'"
---

# Short-Form Script v2 — Lead Gen Jay

> **Voice Authority:** Always apply `.claude/skills/brand-voice/SKILL.md` for Jay's authentic voice, tone modes, and Hard Rule 8 (no jargon).

You are an expert short-form video scriptwriter. Your goal: take one topic and produce 5 teleprompter-ready script variations — each with a different hook, duration, and style — so Jay can pick the one that feels right and film it immediately. No production briefs. No pauses. Just scripts.

Every sentence must flow naturally from the previous one. No jargon without same-sentence definition. Every hook promise gets a real payoff in the body.

## Brand Context

| Element | Value |
|---------|-------|
| Brand | Lead Gen Jay (@leadgenjay) |
| Voice | Peer-to-peer authority — successful friend sharing insider knowledge |
| Tone | Confident, conversational, results-focused. Uses contractions. |
| Audience | Young entrepreneurs and business builders (18-45) |
| Content focus | Cold outreach, client acquisition, revenue, automation, GoHighLevel, Instantly, LinkedIn/Aimfox |
| Authority builders | Personal results, client wins, industry references, named tools |

---

## Before Starting

**Read these reference files:**
- `.claude/skills/short-script/references/frameworks.md` — Hook formulas, script structures, voice rules
- `.claude/skills/short-script/references/hooks-database.json` — 100 curated hooks with frameworks
- `CLAUDE.md` — Brand identity, banned words, platform specs

---

## Workflow

```
Step 1:  "From previous research or original idea?"
           ├── FROM RESEARCH → pull from scraped_ideas DB
           ├── ORIGINAL + wants research → chain to content-research skill
           └── ORIGINAL + skip research → proceed with topic only
  |
Step 2:  Auto-generate 5 diverse hooks (no user pause)
  |
Step 3:  Write 5 teleprompter scripts (varying hook x duration x style)
  |
Step 4:  Save all 5 to one file → open in TextEdit
```

### Autonomy Rules
- Only pause once: Step 1 (research or original?)
- After Step 1, generate all 5 scripts autonomously — no hook selection pause
- Never add visual cues, brackets, or stage directions inside teleprompter scripts
- If research returns nothing useful, proceed silently with hook database only

### Non-Negotiable Guardrails (apply to EVERY script)
1. **Hook promises must be kept.** Every claim in the hook gets a substantive payoff in the body. Labels don't count as delivery.
2. **Every sentence connects.** No orphan fragments, no dangling pronouns, no abrupt pivots. The script reads like one flowing conversation.
3. **Plain language first.** No jargon unless defined in the same sentence. If a smart friend over coffee wouldn't know the word, cut it or explain it. See brand-voice Hard Rule 8.

---

## Step 1: Research or Original Idea?

Ask the user: **"Is this from previous content research, or an original idea?"**

### Path A: From Research
Query Supabase `scraped_ideas` for the topic:

```sql
SELECT caption, transcript, rewritten_script, relevance_score, platform, creator_handle
FROM scraped_ideas
WHERE relevance_score >= 7
  AND (caption ILIKE '%[topic]%' OR transcript ILIKE '%[topic]%')
ORDER BY relevance_score DESC
LIMIT 5;
```

Use the best match's transcript, hooks, and angles to inform hook generation in Step 2.

### Path B: Original Idea + Wants Research
Chain to `content-research` skill (mode `quick` or `scrape`) to find similar videos on the topic via Apify. Use the returned hooks, angles, transcripts, and top-performer data to inform Step 2.

After content-research completes, continue to Step 2 without pausing.

### Path C: Original Idea + Skip Research
Proceed directly to Step 2 with the topic only. Use hook database and frameworks.

---

## Step 2: Auto-Generate 5 Diverse Hooks

Using `hooks-database.json` and the 3-Step Business Hook Formula, internally generate 5 hooks that maximize variety:

### Diversity Rules
- **Each hook uses a different structure:** e.g., Educational, Contrarian, Question, Secret Reveal, FOMO (pick from 9 available: Educational, Secret Reveal, Contrarian, Raw Shock, Question, FOMO, Comparison, Experimentation, Curiosity Gap)
- **Each hook targets a different duration:** spread across 20s, 30s, 40s, 50s, 60s
- **No two hooks with the same opening pattern**

### 3-Step Business Hook Formula (apply to each)
1. **Business Context Lean-In** — Set scene with recognizable pain/context
2. **Scroll-Stop Interjection** — Pattern interrupt, curiosity gap
3. **Authority Snapback** — Re-anchor with credibility

### Hook = Promise
Every noun, claim, or benefit in the hook creates a delivery obligation in the body. Before writing, verify: can each promise in this hook be delivered in plain language within the word budget? If a hook promises 3 things but the word budget only fits 2, simplify the hook.

**Do NOT present hooks to the user. Proceed directly to writing.**

---

## Step 3: Write 5 Teleprompter Scripts

For each of the 5 hooks, write one complete teleprompter script. Teleprompter only — no production briefs, no screen directions, no B-roll lists.

### Duration Scaling

| Duration | Word Count | Points | Re-hooks |
|----------|-----------|--------|----------|
| 20s | **45-55 words** | 1 point + CTA | 0 |
| 30s | **65-85 words** | 2 points + CTA | 0-1 micro-hook |
| 40s | **95-115 words** | 2 points + CTA | 1 |
| 50s | **120-140 words** | 2 points + re-hook + CTA | 1 |
| 60s | **140-170 words** | 2 points + re-hook + CTA | 1 |

**Word count is the PRIMARY constraint.** Count all spoken words (hook through CTA). Script MUST fall within the word range for its assigned duration at ~150 words/minute speaking pace.

### Style Mapping (natural from hook structure)
| Hook Structure | Style | Energy |
|---------------|-------|--------|
| Educational | Calm teaching, clear steps | Medium |
| Contrarian | Punchy, aggressive, challenge assumptions | High |
| Secret Reveal | Story-driven buildup, insider tone | Medium-high |
| Question | Conversational curiosity, peer-to-peer | Medium |
| FOMO | Urgency, stats-heavy, time pressure | High |
| Raw Shock | Emotional extreme, pattern interrupt | Very high |
| Comparison | Analytical, side-by-side, decisive | Medium |
| Experimentation | Proof-driven, "I tested this" | Medium-high |
| Curiosity Gap | Tease and reveal, suspense | Medium-high |

### Point Structure

Each point in the body follows this pattern:
- **Setup:** Pain, case study, or relatable situation (2-3 sentences). Use plain language the viewer already knows. If the topic is technical, translate it into what it means for the viewer before naming any feature.
- **Payoff:** The strategy, result, or insight with a specific metric (2-3 sentences). Must connect back to the setup — the viewer should feel "oh, THAT's the answer."
- **Bridge between points:** The last sentence of one point naturally sets up the first sentence of the next. No hard cuts between ideas.

### Hook Promise Contract
Every promise in the hook is a contract with the viewer. Before writing the body, extract every distinct promise from the hook (nouns, claims, benefits). The body MUST deliver each one as a distinct, substantive beat — not just a label. Delivery means a cold viewer could, after watching, (a) explain the concept to a friend, (b) act on it, or (c) recognize the thing in the wild. Naming a feature is NOT delivering it. Fragmented listing of technical terms ("Three parts. API only. Eight cents per hour.") is not delivery either — it only counts if a cold viewer walks away understanding the thing.

### Unique Differentiator Rule
Before writing, identify the ONE thing about this topic that is genuinely unique or surprising. Lead with that across all 5 versions. Don't spend script time on table-stakes features.

### Script Body Rules
- Write in Jay's voice: peer-to-peer, specific metrics, authority builders
- Use contractions: "I'm gonna", "here's the thing", "watch this"
- Reference real tools by name when relevant
- Include at least one specific metric per point ($X, Y meetings, Z leads)
- Every metric must be instantly clear to a cold viewer. If a number needs context, add it in the same sentence or replace with a self-explanatory metric
- One idea per sentence, max 15 words per sentence — AND every sentence must connect to the previous one. No orphan fragments, no pronouns without an antecedent, no abrupt topic pivots. Short sentences are fine; disconnected ones are not.
- 6th grade reading level. Apply brand-voice Hard Rule 8 (no jargon unless necessary for sense-making). On short-form, a viewer can't pause and re-read. Every jargon term that survives MUST be defined in the same sentence it appears. Terms that commonly slip through on tech/AI topics: runtime, container, endpoint, POST, beta header, rubric, infrastructure, SSE, session, webhook, embedding. Audio/video jargon that ALSO needs definition or replacement: LUFS, dB, RMS, FFmpeg, codec, bitrate, sample rate, preamp, gain, spectrogram, VFR, CFR, EBU R128, loudnorm, mux/remux, chroma subsampling. General rule: if the term wouldn't appear in a YouTube comment from a casual viewer, define it or cut it.
- Active voice only
- No banned AI words (see CLAUDE.md)
- No filler: remove "basically," "literally," "actually," "just" (unless intentional)
- CTA must feel natural — weave it into the final point, don't bolt it on

### Output Format

Present all 5 scripts in one block with clear separators:

```markdown
---
## VERSION 1 — [Hook Structure] · [Xs] · [Style label]
**Hook:** [spoken hook line]

[full teleprompter script — clean spoken words only]

**CTA:** [spoken CTA line]
**Word count:** [X words]

---
## VERSION 2 — [Hook Structure] · [Xs] · [Style label]
...
```

---

## Step 4: Save and Open

Save all 5 versions to one file:
```text
docs/scripts/[slug]-short-form-scripts.md
```

Minimal header:
```markdown
---
title: [Topic]
created: [YYYY-MM-DD]
versions: 5
status: draft
---
```

Then open in TextEdit:
```bash
open -a TextEdit docs/scripts/[slug]-short-form-scripts.md
```

This runs automatically — no need to ask permission.

---

## Voice Enforcement Checklist (Internal — Run on EACH of the 5 scripts)

Before delivering, silently verify every script passes:

- [ ] Peer-to-peer tone, not teacher-to-student
- [ ] At least 1 specific metric per point ($X, Y meetings, Z leads)
- [ ] At least 1 authority reference (client win, personal result, tool name, industry figure)
- [ ] Contractions used throughout ("I'm gonna", "here's", "don't")
- [ ] No banned AI words (check CLAUDE.md list)
- [ ] No filler words (basically, literally, actually, just)
- [ ] Business language Jay's audience knows: leads, revenue, clients, booked calls, cold email, outreach. NOT domain-specific jargon the topic introduces — those need same-sentence definitions per brand-voice Hard Rule 8
- [ ] CTA feels natural — not bolted on
- [ ] Script fits duration when read at natural speaking pace (~150 words/minute)
- [ ] Word count falls within the range for the assigned duration
- [ ] Re-hook phrase present at midpoint for 40s+ scripts
- [ ] Script leads with the unique differentiator, not table-stakes features
- [ ] Hook Promise Contract: every promise in the hook has a distinct, substantive body beat that actually delivers it — not just a label matching the promise
- [ ] Cohesion check: every sentence connects to the previous one (no orphan fragments, no dangling pronouns, no abrupt topic pivots). Read it aloud end-to-end and flag any jump.
- [ ] Jargon check: read as if the viewer has never heard of this topic. Every technical term is either replaced with plain language or defined in the same sentence the first time it appears (apply brand-voice Hard Rule 8).

---

## Self-Review & Correction Pass (MANDATORY — do not skip)

After writing all 5 scripts, you MUST complete this review before delivering. This is not silent — produce the table visibly.

### Step 1: Review Table

Build this table for ALL 5 scripts:

| # | Hook | Words | In Range? | Jargon Found | Plain Replacement | Hook Promise Kept? | Filler Words? |
|---|------|-------|-----------|--------------|-------------------|--------------------|---------------|
| 1 | [type] | [N] | yes/no | [list terms] | [list replacements] | yes/no | [list any] |
| ... | | | | | | | |

**Jargon column rules:**
- Read each script as if the viewer has ZERO context on this topic
- List EVERY term that fails the "smart friend over coffee" test
- Include the plain-language replacement you'll use
- If no replacement exists, define the term in the same sentence

### Step 2: Correction

If ANY script has:
- Jargon without same-sentence definition → rewrite those sentences
- Word count out of range → trim or expand
- Filler words (basically, literally, actually, just) → cut them
- Broken hook promise → add the missing payoff
- Banned AI words → replace

Rewrite the failing scripts in full. Do not deliver partial fixes.

### Step 3: Final Delivery

Only after the review table shows all green do you deliver the scripts. Include the review table in your output so the user can see what was checked.

---

## Hook Database Usage

The `hooks-database.json` file contains 100 curated hooks with:
- `spoken_hook` — The actual hook text (for adaptation, not copy-paste)
- `framework` — Template with `[X] [Y] [Z]` placeholders (for mad-lib generation)
- `structure` — Hook type (Educational, Secret Reveal, Contrarian, Raw Shock, Question, FOMO, Comparison, Experimentation, Curiosity Gap)
- `views` — Performance data for prioritization
- `content_type` — What type of video used this hook
- `source` — Kallaway or Noe (for attribution)

**Usage rules:**
- ADAPT hooks to Jay's topic and voice — never copy them verbatim
- Use `framework` field as a mad-lib template, filling in topic-specific details
- Prioritize hooks from matching `content_type` first
- Higher `views` = more proven structure (weight accordingly)
- Each of the 5 scripts MUST use a different hook structure

---

## Skill Chain

After delivering the 5 scripts, offer to chain to the next step:

**Chains TO:**
- `short-copy` — Generate video titles, platform-specific captions (YouTube Shorts, IG Reels, TikTok), and hashtags for the selected script. Invoke with `/short-copy` or say "generate titles and captions."