sxsphinxstack

Skills / Video script

Video script

Script a video they will actually record. Hook, beats, words they'd say out loud, rehearsed once. Use when they say "help me script a video", "I need to record a demo", or want to make a short or walkthrough.

Use this skill. Nothing to install.

Open in ChatGPT Open in Claude prefilled and ready; just hit send
for Gemini, Copilot, or Cursor: paste it, then say "use the video script skill"

Don't have an agent?  ·  Raw file: skills/video-script.md

Watch the first two minutes

This is how a session goes, including the trick that fixes the robot voice. Click through it.

​(pastes the skill) I need a demo video for my project but every time I write a script and read it, I sound like a hostage tape
That's what happens when prose written for the page gets read out loud, so we won't write prose. Explain your project to me right now, in chat, casually, like you'd tell a friend at lunch. Don't make it good, just tell me.
ok so basically my bus stop's schedule lies, so I built a page that shows where the bus actually is. you look at it before you leave the house and you just know if you can chill or you need to run
"You know if you can chill or you need to run." That's your voice, and that line goes in the script. I've saved what you wrote; it's the raw material. Now we shape it into beats: a hook, three or four points, a close. First, how long is this video?
I was thinking 5 minutes? there's a lot to show
First video gets 60 to 90 seconds, which at speaking pace is about 150 words. Shorter than you think. One video, one point: when it ends, what does the viewer do? If your answer contains "and", we cut it in half.
they open the site and check their own stop I guess. fine. what's the hook, "hey guys welcome back"?
Banned, along with "in this video I'm going to". The first five seconds show the thing itself: your screen, the bus moving on the map, and you saying why it exists. Write me three different hooks and we'll say them out loud to pick one. The hook and the close get scripted word for word; everything else stays bullet beats.

Scripted example of a real session.

I freeze the second the camera is on

The skill plans for that instead of pretending confidence will show up. The script is bullet beats in words you already said in chat, so there's nothing to memorize and forget. You rehearse the whole thing out loud exactly once, timed, before recording. And the first take is a throwaway by design, which removes most of its power to scare you. Two or three takes, keep the best, trimmed at both ends, done. Editing ambitions are for video two.

What you end up with

A recorded, uploaded video and the beat sheet that made it. The beat sheet looks like this.

Example
bus42 demo  ·  75 seconds
HOOK, WORD FOR WORD 1
"This is the 42 bus, live. The schedule at my stop says it comes at 8:10. It's 8:14 and it's still six blocks away."
Screen: the map, bus marker moving
BEATS 2
Why I built it: the schedule lies, I kept missing it
Screen: photo of the posted schedule
What you see: bus position, arrival guess for your stop
Screen: click a stop, guess appears
The honest part: rush hour guesses run late, working on it
Screen: stay on the map
CLOSE, WORD FOR WORD 3
"It's at busmap.example.dev. Try it with your own stop tomorrow morning."
Screen: the URL, big, then cut
  1. The hook is the thing itself. No greeting, no channel talk. A concrete moment a viewer recognizes, on screen within five seconds, scripted word for word because nerves eat first lines.
  2. Beats are prompts to talk over. One line of what gets said, in words from your own chat explanation, next to what's on screen while you say it. A demo where the screen lags the voice loses the viewer, so shots are planned with the lines.
  3. The close is one action, then stop. No fade-out thanking paragraph. Say where it is, say what to do, end the video.

Questions people actually ask

Is this free?

The skill is free, no account, no catch. The agent you paste it into may or may not be: Gemini has the widest free tier, Copilot's free tier is enough to try it, ChatGPT and Claude free plans can also run it in a chat. Recording uses the phone or screen recorder you already have.

Will the script sound like AI wrote it?

It can't, structurally: the script's raw material is you explaining the thing casually in chat, and the beats reuse your phrasing. The agent shapes and times; it doesn't write lines for you to recite. Any line you stumble on or would never say in conversation gets rewritten by you, out loud, on the spot.

Do I need a good camera or mic?

No. Phone camera or the screen recorder built into your computer. The skill's one production rule is that audio matters more than video: quiet room, mic close. Fancy gear is nowhere on the path for video one.

How long should my first video be?

60 to 90 seconds. At a natural speaking pace of roughly 140 words a minute, that's a shorter script than most people expect. If the rehearsal read runs over, you cut beats rather than talk faster.

Does the session really end with recording?

Yes, or at worst with you pressing record as it ends. A script that never meets a camera taught you nothing about making videos. First take is a throwaway by design, then two or three real takes, keep the best, upload it where it was needed.

Where to go from here

Video recorded? Finish the pipeline:

Edit a short Publish to YouTube Announce it without cringe

Want the next one to look and sound better?

Record clean audio Make a thumbnail Captions and subtitles
Curious? Read the full skill — the exact instructions your agent gets 
---
name: video-script
category: write
description: Script a video they will actually record. Hook, beats, words they'd say out loud, rehearsed once. Use when they say "help me script a video", "I need to record a demo", or want to make a short or walkthrough.
---

# video-script

Help someone script a short video they will record —
a project demo, an intro, an explainer, a short. A script for the
page and a script for the mouth are different animals: this one has
to survive being said out loud on camera by them, in their voice.
They write it by talking; you structure and time it. The session
ends at the recording, or at worst with them pressing record as you
leave.

## Scope it down

- What is the video, who is it for, where will it live (YouTube,
  a short, a course application, a repo link)?
- One video, one point. Ask: "when it ends, what does the viewer
  know or do?" If the answer has an 'and', cut it in half.
- Length: 60–90 seconds for a first video. Roughly 140 spoken
  words per minute — a one-minute script is shorter than they
  think.

## Script by talking

Start from speech rather than polished prose so the script sounds natural when
the person reads it aloud:

1. They explain the thing to you in chat, casually, as if to a
   friend. Save that text — it is the script's raw material and it
   is already in their voice.
2. Together, shape it into beats: hook, then 3–5 points, then the
   close. Each beat is one line of what gets said, in their words
   from step 1.
3. The hook is the first five seconds and decides everything. Ban
   "Hey guys, welcome back" and "In this video I'm going to."
   Start with the thing itself: the problem, the demo, the result
   on screen. Have them try three hooks and pick one.
4. The close says what to do next — one action ("link in the
   description", "try it at this URL") — and stops. No fade-out
   thanking paragraph.

Keep the script as bullet beats, not full sentences to memorize.
Exception: the hook and the close get written word-for-word,
because those are the lines nerves eat first.

## Say it before you shoot it

- They read the whole script aloud to you once, timed. Over
  budget: cut beats, never talk faster.
- Any line they stumble on or would never say in conversation gets
  rewritten on the spot — by them, out loud, then written down.
- If the video shows a screen, list the shots next to the beats:
  what is on screen while each line is said. A demo where the
  screen lags the voice loses the viewer.

## Record

Recording is part of the skill. Phone camera or screen recorder
they already have; audio matters more than video, so quiet room,
mic close. First take is a throwaway by design — that removes the
fear of it. Take two or three, keep the best, done. Talk them out
of editing ambitions on video one: one good take, trimmed at both
ends, uploaded.

## Done

- The timed script has one audience, one point, and checked on-screen actions
- The person read it aloud, revised any unnatural line, and recorded the video
- The final video is uploaded or embedded where needed, and the script beats
  are saved for the next one

If the
video announces something they made, the social-post skill handles
telling people about it.