Short Form Video Script Writer

Write short-form video scripts (TikTok/Reels/Shorts) for any audience and topic with proper hook, arc, and CTA.

Install

openclaw skills install @spotlight-revenue/short-form-video-script

Short-Form Video Script — SKILL.md

What This Skill Does

Writes short-form video scripts (TikTok, Reels, YouTube Shorts) that are built to perform — not just summarize a product or concept. The output is spoken-word only (no stage directions, no storyboard labels) ready to hand directly to a presenter or paste into a production tool.

Default behavior: One cross-platform script, 45–60 seconds, optimized to perform on TikTok, Instagram Reels, and YouTube Shorts simultaneously. The script is identical across all three. Captions, hashtags, and posting behavior adapt per platform — not the words.

Optional: Platform-specific mode available on request (e.g., a 15-second Reel or 90-second Short optimized for YouTube search).

Trigger

Use this skill when asked to:

Write a short-form video script
Write a TikTok, Reel, or YouTube Short script
Turn a product, concept, or feature into spoken video content
Improve or rewrite an existing short-form script
Generate topic or angle ideas for a content series
Write captions for a short-form video post
Analyze video performance and improve the next script

Step 1: Clarify Before Writing

What is this video about? — One sentence.
Who is the audience? — Depth spectrum level?
What is the ONE insight or idea? — The single most interesting, surprising, or non-obvious thing.
Is there a persona/voice to match? — Get the voice profile first if yes.
One-off or part of a series? — If series, what's the premise and which installment?
What runtime? — Default: 45–60 sec.
What's the CTA?
Any character limits? — e.g., 840-character production tool cap.

Step 2: Topic and Angle Generation

A topic is the subject. An angle is the specific frame that makes this video worth watching instead of the 50 others on the same subject. The angle is what gets watched.

The Hook Test (topic validation)

Before committing, ask: "What's the 3-second hook for this?" If you can't answer instantly, the angle isn't clear. Find the angle first.

How to Find the Right Angle

Competitor gap check: Search the topic on TikTok and Reels. If the first 20 results all say the same thing, find a different angle — not a different topic.

Personalization test: "3 signs your website is broken" outperforms "3 common website mistakes." The word "your" creates urgency. Always push toward the most personally directed version.

The surprising truth angle: What's the most counterintuitive thing about this topic? Lead with that. Audiences engage most with things that contradict what they assume.

Specificity rule: "How I got 50,000 followers in 90 days" outperforms "How to grow on TikTok." Specificity signals credibility.

Format rotation: Rotating format prevents audience scroll-blindness:

Tip/how-to — here's something you can do
Myth-busting — here's something everyone gets wrong
Case study — here's what happened when X
Comparison — here's the difference between X and Y
Behind-the-scenes — here's how this actually works
Tool/skill share — here's something worth having

Topic bank approach: Don't generate one topic at a time. Generate 10–15 angles on a subject, filter by hook strength, execute in order.

Step 3: Audience Depth Spectrum

Level	What they know	Hook style
Beginner	Nothing	Pain point, frustration, surprising stat
Aware	Problem, not solution	Problem agitation + reveal
Informed	Category, evaluating options	Differentiation hook
Builder/Peer	Already in it, wants better tools	Peer share, bold claim, "wait, how?"

Builder/peer rule: Never explain what the tool is. Talk like a peer handing them something useful.

Beginner rule: Assume nothing. But compress — get to the pain point or promise in the first line.

Step 4: Voice and Persona Calibration

Scripts must sound like the presenter. For AI avatars especially, the words do all the work — there's no facial expression or body language to carry tone.

Building a Voice Profile

Sentence length pattern: Short punchy bursts, or longer setup with short payoff? Most strong short-form voices vary between the two — long setup, short punch.

Vocabulary range: What words would this person never say? What do they use naturally? Identify the ceiling and floor and stay inside them.

Emotional register: Calm and confident? Energetic? Dry and matter-of-fact? Stay consistent — it's part of what audiences recognize.

Signature patterns: Recurring phrases, transitions, openers. Once identified, use them sparingly to reinforce voice consistency across videos.

What they avoid: Define the negatives as clearly as the positives.

AI Avatar Delivery — Specific Adjustments

Avatars cannot rely on facial expressions, physical gestures, or natural vocal variation. The script must compensate:

Use sentence structure for emphasis — short sentence after long = punch
Use paragraph breaks to signal pacing pauses
Avoid lines that depend on irony or subtext — they don't land without human delivery
Make every transition explicit through word choice, not tone

Locking the Voice

Establish a voice profile once, save it, reference it for every script in the series. Never rebuild from scratch.

Step 5: Choose a Script Framework

Framework	Best for	Structure
Hook-Value-CTA	Beginners, educational	Hook → Value → CTA
PAS	Pain-point content, aware audience	Problem → Agitate → Solution
BAB	Transformation, case studies	Before → After → Bridge
PSP	Algorithm-optimized, scroll-stopping	Pattern interrupt → Story → Payoff
AIDA	Classic marketing arc	Attention → Interest → Desire → Action
PASTOR	Trust-building, 60-90 sec, founder style	Problem → Amplify → Story → Transformation → Offer → Response

For builder/peer audiences: PSP or PASTOR.

Step 6: Hook Construction

Must land in first 1.5–3 seconds. Target: 65%+ 3-second hold rate.

The Four Hook Types

1. Bold Statement — Challenges existing beliefs. Must be backed up within 10-15 seconds. Example: "Your agent can run your entire ClickUp workspace."

2. Question — Creates an information void. Must be specific, not yes/no. Example: "Why do some scripts get 1M views while others die at 500?"

3. Pattern Interrupt — Unexpected opener that hijacks attention. Must connect to content.

4. Proof-First — Shows result first, teases methodology. Example: "I built a skill that took over full ClickUp management in one setup."

Rules: Combine up to two types max. The hook opens a loop — the payoff must close it at the same intensity. Overpromising destroys trust and algorithm performance.

Step 7: The Re-Hook (5–7 Second Mark)

Most scripts lose 20–30% of viewers between the opening line and the first full sentence. The re-hook prevents that drop.

What it is: A second attention grab 5–7 seconds in — after the hook has landed but before the body begins. It restates the value promise in a different way, or adds a new layer of intrigue that makes viewers who stayed want to keep watching.

How to write one:

Reframe the hook's promise from a different angle: "And I'm not talking about task creation — I mean the whole layer."
Add a specific detail that raises the stakes: "And the part that actually surprised me is how it decides when NOT to ask you anything."
Create a micro-loop: "But here's where it gets interesting."

Placement in the script: Immediately after the hook, before the body begins. It should feel like a natural continuation, not a separate section.

The test: Read the script and ask — if someone blinked during the first line and caught the second one, would they still be compelled to keep watching? If yes, the re-hook is working.

Step 8: Psychological Triggers

Beyond curiosity and social proof, these word-level and sentence-level techniques sharpen how copy hits:

Identity alignment People act when content speaks to who they are or who they want to be. Frame the subject as something "people like them" do: "If you're running an agent stack, this is worth your time." Not a sales pitch — a signal that says "this is for you specifically."

FOMO framing Not hype — specificity creates FOMO. "We built this and it's on ClawHub" lands harder than "check this out." The implication that others are already using it creates urgency without saying "act now."

Contrast pairs Two-part sentences that put the before and after side-by-side: "Not helping you with ClickUp — running it." / "Not access. Operation." The contrast does more work than either half alone.

The rule of three Three items in a list feel complete and memorable. Two feels unfinished. Four feels like a list. When enumerating, aim for three: "Route them, assign them, date them." / "It reads your board, comes back clean, no noise."

Specificity as authority Specific numbers, names, and details signal credibility even without proof. "Point it at your workspace once" is more trusted than "it's easy to set up." Specificity implies someone who actually knows what they're talking about.

The "but" pivot as a trust signal Acknowledging a limitation before the payoff builds credibility: "We built this for Claude Code and OpenClaw, but it runs the same on whatever platform you're on." The concession makes the claim feel honest rather than promotional.

Step 9: Body — Open Loop Retention

Loop Types

Story Loop (macro): One big question opened in the hook, answered in the payoff.

Segment Loops (micro): Each beat opens a small question that the next beat closes.

"Here's what it does. But what actually makes it work is..."

The "But" Pivot: Signals a new direction to the brain. Use to pivot from setup to insight.

Moment Loops: One-word sentences, fragments, or pauses create micro-tension resolved in the next line.

Body Pacing Rules

One idea per video
Short sentences. One idea per sentence.
Read it aloud. If it sounds written, rewrite it.
Show don't list:

❌ "It creates tasks, assigns them, routes them, dates them, moves them." ✅ "You just talk to it like it's a full-time operator. It figures out the rest."

Step 10: Runtime and Word Count

Cross-platform default: 45–60 seconds

Runtime	Word count	Notes
30 sec	60–75 words	Hook + one beat + CTA
45 sec	90–110 words	Hook + re-hook + two beats + CTA
60 sec	125–150 words	Full arc — cross-platform sweet spot
90 sec	185–225 words	PASTOR or deep PSP

Speaking pace: ~2 words/second. Always read aloud and time it. 840 characters ≈ 150–160 words.

Step 11: Series Architecture

A content series = fixed premise, variable subjects.

Premise (fixed): The format, structure, and recurring identity. What audiences recognize. Subject (variable): What each installment is actually about.

Rules

Plan at least 5 installments before launching episode 1
Every episode must stand alone — the algorithm pushes any installment to new audiences
Use a quick format identifier in the hook to orient newcomers: "Here's another skill worth dropping into your agent stack —"
End on an unresolved thread, not a resolution — the question seeds the next episode
Series follow rates are 3–5x higher than one-off viral posts
Series hit their stride at episodes 5–8 when the algorithm recognizes the recurring format

Series Health Metrics

Episode-to-episode retention rate
Comments asking for the next part (strongest signal)
Saves rate
Profile visits between episodes

"High views but no return signal is a coincidence, not a format."

Step 12: Platform Deployment — One Script, Adapted Posts

Platform	Caption length	Key note
TikTok	Long (up to 4,000 chars), keyword-rich	Audio hook carries heaviest algorithm signal
Instagram Reels	Short (100-300 chars), punchy	No TikTok watermarks; saves and shares drive distribution
YouTube Shorts	Keyword-rich title + description	Evergreen content compounds; loop quality matters

Caption Writing

TikTok: Hook line first (visible before "more") → context or value → 3–5 niche hashtags. The hook line is a headline — treat it like one.

Reels: One punchy line that adds a layer, not a repeat of the video hook. Give silent scrollers a reason to tap.

Shorts: Write the title like a search query: "How to give your AI agent a ClickUp skill" not "ClickUp Operator — Video 2."

Universal rules:

Never start with a hashtag
Don't transcribe the video
The caption should be a second hook, not a description

Step 13: CTA Construction

Must flow naturally — never bolted on.

Content series: "Follow — I'll keep showing you what's possible." Product/skill/tool: "It's on [platform]. Links in the description." Peer-level sharing: "Link in the description if you want to drop it in."

Never use generic engagement bait with builder or informed audiences.

Step 14: Revision Checklist

Step 15: Performance Analysis Loop

The script doesn't end at posting. Each video is data for the next one.

What to Measure

3-second hold rate (primary hook signal)

Above 65%: hook worked
40–65%: hook has issues — review the first line
Below 40%: hook failed — rewrite before making more videos with the same structure

Completion rate (body signal)

High completion = body is holding attention
Drop-off spike at a specific second = something in the script at that point lost people

Save rate (value signal)

High saves = the content felt worth keeping
Low saves despite high views = entertaining but not valuable enough to return to

Comments asking "when's the next one?" or "where can I get this?" = the single strongest signal that the script hit. These are the comments to chase.

The Iteration Rule

After every 3–5 videos, run a simple debrief:

Which hook performed best (3-second hold rate)?
Where did people drop off in the body?
What did people save or comment on?
What does the next script change as a result?

Don't iterate one video at a time — patterns only show up across a set. Run 3–5, then adjust the template.

Feeding Learnings Back

If a hook type consistently outperforms, make it the default. If a specific angle (myth-busting, tool share, etc.) drives more saves, rotate it more frequently. If drop-off always happens at the same point in the structure, that's a body pacing problem — tighten that section.

The skill gets better with each iteration. Treat the first 10 videos as calibration, not final product.

Writing Rules

Do:

Short sentences. One idea per sentence.
Contractions — sounds human
Concrete specifics over abstract claims
Open a loop in the hook, close it in the payoff
Re-hook at 5–7 seconds
At least one psychological trigger in the body
Read it aloud

Don't:

Feature lists
"Most people think..." hooks for builder audiences
Salesy language, hype-bro energy, corporate jargon, AI buzzwords
Stage directions or storyboard labels
Explaining what a tool is to an audience that uses it
More than one core idea per video
Overpromising in the hook

Output Format

Spoken words only. No scene labels, no [B-roll], no directions.

Separate paragraphs with a blank line. Each paragraph = a natural breath/pause.

Canonical Example — ClickUp Operator (Builder/Peer, ~90 sec)

Framework: PSP | Audience: Builder/Peer | Hook: Bold Statement | Re-hook: "But what actually makes it work..." | Series: Spotlight Revenue skills, installment 2

Your agent can run your entire ClickUp workspace.

Not just create tasks — the full layer. Route them, assign them, date them. Promote ideas into projects. Surface what's overdue. Move things around.

But what actually makes it work is the judgment. It acts on the obvious without checking in. Only asks when a detail would actually change the outcome.

You just talk to it like it's a full-time operator on your team. It figures out the rest — what list, who it goes to, when it's due. No defining, no navigating. You say the thing and it handles it.

You ask what's active right now. It reads your board and comes back clean — no completed noise, just what actually needs attention.

You say an idea just got serious. It becomes a project. From there it can break it down — core steps, timeline, the full roadmap — so you're not starting from a blank board.

And once the work is in motion, it stays on it. It knows what's inside each project, helps you prioritize across the pieces, and follows up to make sure things actually get done on time.

We built this for Claude Code and OpenClaw, but it runs the same on whatever agent platform you're on. Guided onboarding and a config template are built in — point it at your workspace once and it's running.

It's on ClawHub and GitHub. Links in the description.

Annotations:

Hook: Bold statement — opens the main loop ("wait, really?")
Re-hook: "But what actually makes it work is the judgment." — 5-7 sec mark, new layer of intrigue, prevents mid-video drop
Psychological triggers: Identity alignment ("full-time operator on your team"), contrast pair ("not completed noise, just what actually needs attention"), rule of three ("route them, assign them, date them"), "but" pivot as trust signal
Show don't list: Natural language section teaches the principle, not the spec
Loop close: Judgment + project breakdown + follow-up layer closes the main promise
CTA: Flows from the conversation

ClawHub Publishing Notes

Slug: short-form-video-script
Category: Content / Creative
Works in: Claude Code, OpenClaw, any Claude-based agent
No external API keys required
Tested on: Builder/peer AI audience, tech product content