360 Secure SkillHub
360 安全技能中心

Phantra AI: Youtube Script

面向YouTube视频脚本创作:先确认主题、受众、时长与题材,再按题材结构撰写钩子、分段和单一行动号召,以SAY/SEE两栏呈现,并用字数对齐时长;事实须由用户提供,缺数据留占位。用户明确要求时才把SEE列转为分镜场景。

分类:创作
zeropage-onryo
zeropage-onryo
热度 9.94
最近更新 2026-11-03 02:11:13

安装

发给 Agent
帮我安装这个技能:npx secure-skills-cli add zeropage-onryo/phantra-ai-youtube-script
兼容性
通用兼容
联网要求
联网要求未确认
数据出网
可选出网
Metadata
name
youtube-script
description
Writes a YouTube video script in the chat, shaped to its genre (explainer, list, tutorial, review, story), with a hook, sections and one call to action, set out as what is SAID beside what is SEEN; only if the person wants footage does the SEE column go to Phantra as scenes. Writing it spends nothing. Use when someone asks for a YouTube or video script, or an outline. NOT for one scene's prompt (write-scene), scenes with no spoken script (storyboard) or the cover (thumbnails).

YouTube script

A script is writing, done here in the chat. Nothing in this skill costs credits, and no tool is needed until the person asks for footage. The studio makes pictures, not narration: the voice is theirs to record.

1. Ask before writing

  • The topic and the one thing a viewer should leave with.
  • Who it is for, and what they already know.
  • The length in minutes, and the genre (below).
  • What they know that a viewer does not: their numbers, their story, their examples. These are the script; ask for them.

2. The shape, by genre

Every genre opens with a hook and ends with one call to action. The middle differs:

genre the middle
explainer the question, the plain answer, how it works in three parts, the common mistake
list the items in rising order of interest, the strongest held for last
tutorial the finished result, the steps in order, each with its one pitfall
review the verdict, what was tested and how, what is good, what is not, who should buy
story where it started, the turn, what it cost, what changed
commentary the claim being answered, their position, the evidence, the strongest objection

The hook (the first 15 seconds): say what the video delivers and show that it will, before any greeting, channel name or "in this video". One sentence of promise, one of proof.

Between sections: a line that opens the next question, so there is a reason to stay. The end: the payoff of the hook's promise, then ONE thing to do next. Two asks are none.

3. Write it as SAY beside SEE

Two columns, a row per beat. SAY is the words, in their voice: short sentences, written to be spoken. SEE is what is on screen: them to camera, a screen recording, a product close-up, a title card.

# SAY SEE
hook "I ran this for a month. Here's what broke." to camera, mid-gesture; cut to the broken part held up
1 ... ...

A spoken minute is roughly 150 words (the studio's advice, not a rule), so say the word count beside the target length and cut to fit.

4. What is theirs, what is checked

Every fact, number, quote, test result and claim in SAY comes from the person in this conversation. Never invent a statistic, a study, a quote or an experience; where a line needs a fact you do not have, leave a marked gap ("[your number here]") and ask. Never write a review of a thing they have not used, and never write a line as if someone else said it.

5. If they want footage from the studio

Only on their ask. The SEE column becomes scenes; the SAY column stays a voice they record.

  • Each scene is 4 to 30 seconds of picture, rendered one timed shot at a time, and a rendered scene needs at least one of their photos. So the studio suits the inserts (a product close-up, a re-creation, an opener), not ten minutes of them talking.
  • Pick the rows worth a scene with the person, then plan them together in storyboard (one write_scene per scene, filed in one project made with create_project) or write a single one with write-scene.
  • Writing and saving a scene is free. Rendering is priced per scene by quote and runs only on their yes, through approve (price-and-render). Say how many scenes the plan is before quoting any.
  • Finished clips are in library; assemble_clips joins them (no credits) and share_link hands over the file for their own editor, where the voice goes on. Scenes render in a vertical frame; reformat with aspect 16:9 makes a wide copy, padded or cropped, at no credit cost.

6. Keep what works

  • save_recipe keeps their script format (the genre's order, their hook pattern, their sign-off) under a name, so the next script starts from it; recipes lists what they have saved.
  • For the video's description or a caption, score_caption scores a draft against their own posted videos; with too little history it says so.

Do not

  • Never say the studio can narrate, add a voice, write subtitles onto the video or upload to YouTube; none of that is here.
  • Never call a tool that spends while writing a script.
  • Never pad to a length: cut the weakest section instead.