Token导航 LogoToken导航

Agent Skill · 前端设计

educational-video-creator(教育视频创作者)

用于辅助视频生成、动画合成、脚本化剪辑或 Remotion 等视频项目开发。它适合让 Agent 组织镜头、生成素材说明、维护合成代码或排查渲染问题。使用时需要确认分辨率、时长、素材路径和导出格式;涉及外部素材、人物肖像或商业发布时,应先核对版权授权和内容审核要求。

educational-video-creator 是前端设计类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为需要联网;许可证信息还不明确,第三方扫描有未通过项,按安装量在前端设计类的 10,216 个 Skill 里排第 1,254。

总安装 6,703周安装 285GitHub Stars 33下载 2,348核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:educational-video-creator(教育视频创作者)
来源仓库:https://github.com/skindhu/skind-skills
仓库路径:skills/educational-video-creator
安装命令:
npx skills add https://github.com/skindhu/skind-skills --skill educational-video-creator
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/skindhu/skind-skills --skill educational-video-creator

这个 Skill 是做什么的

前端设计类 Skill 关注页面结构、视觉规范和组件写法,输出通常是可运行的界面代码。

支持 Remotion 框架生成类 Kurzgesagt 风格的教学视频项目。可协助镜头编排、素材说明编写及合成代码调试等全流程开发任务。需配合 remotion-best-practices 技能完成技术环境配置后方可使用。涉及人物肖像或商业发布时,须提前核查版权授权与内容审核规范。

educational-video-creator 属于前端设计类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/educational-video-creator

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/skindhu/skind-skills --skill educational-video…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

educational-video-creator 怎么安装?

在终端执行 npx skills add https://github.com/skindhu/skind-skills --skill educational-video-creator。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

educational-video-creator 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

educational-video-creator 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

educational-video-creator 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

educational-video-creator 用的人多吗?

公开数据是累计安装 6,703 次,近一周安装 285 次,下载 2,348 次,GitHub 33 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

educational-video-creator 安全吗?

第三方扫描有未通过项,风险较高。除非你确认过源码,否则不建议直接安装到常用环境。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理前端设计类任务、又不想每次都手写提示词的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

前端设计类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
frontend-design

前端设计 · Codex、Claude

8,823,360126,072只读MIT通过
vercel-composition-patterns

Vercel 组合模式 · Codex、Claude

3,858,81625,893需要联网MIT通过
vercel-react-native-skills

Vercel React native skills 前端 · Codex、Claude

2,468,66425,878只读MIT通过
adapt

adapt 搜索 · Codex、Claude

2,027,04023,125操作浏览器未明确通过
bolder

bolder 分析 · Codex、Claude

2,004,28823,130只读未明确通过
educational-video-creator

教育视频创作者 · 本页

6,70333需要联网未明确未通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Educational Video Creator

Create professional educational videos with Kurzgesagt/回形针 visual style using Remotion.

Prerequisites

This skill requires remotion-best-practices for Remotion technical implementation.

Check and install (MUST complete before Phase 1):

# Check if already installed, install if not
npx skills list 2>/dev/null | grep remotion-best-practices || npx skills add https://github.com/remotion-dev/skills --skill remotion-best-practices

Once installed, read the remotion-best-practices skill to load Remotion API details into context. This is essential — without it, Phase 4 code will have incorrect Remotion API usage.

External dependencies (needed for Phase 4.5 audio generation):

# ffmpeg (provides ffprobe for audio duration measurement)
brew install ffmpeg        # macOS
# apt install ffmpeg       # Linux

# edge-tts (Python TTS engine for narration generation)
pip install edge-tts

Project Setup

This skill creates videos in a dedicated remotion_video subdirectory within the current workspace.

First-time setup (recommended — non-interactive):

# Create video project directory
mkdir -p remotion_video
cd remotion_video

# Initialize without interactive prompts
npm init -y
npm install remotion @remotion/cli @remotion/google-fonts @remotion/transitions react react-dom
npm install -D typescript @types/react

# Create minimal project structure
mkdir -p src public/audio/narration

Then create the required entry files:

  • src/Root.tsx — Main composition registry
  • src/index.ts — Remotion entry point with registerRoot(Root)
  • remotion.config.ts — Remotion configuration
  • tsconfig.json — TypeScript config

Alternative (interactive — may block in automated environments):

mkdir -p remotion_video && cd remotion_video
npx create-video@latest --blank
npm install
Note: npx create-video may prompt for project name, package manager, etc. In CLI/automation contexts, use the non-interactive method above to avoid blocking.

Existing project:

cd remotion_video
npm install  # Ensure dependencies are installed

Project structure:

your-workspace/
├── remotion_video/           # Video project root
│   ├── src/
│   │   ├── Root.tsx          # Main composition registry
│   │   └── YourVideo/        # Video-specific components
│   │       ├── index.tsx
│   │       ├── scenes/
│   │       └── components/
│   ├── public/               # Static assets
│   └── package.json
└── ... (other workspace files)

Quick Start

  1. Setup project → create remotion_video directory if needed
  2. Gather requirements → confirm topic, audience, duration
  3. Write script → complete narrative with story arc and pacing
  4. Create storyboard → break script into visual scenes with animation specs
  5. Design visuals → apply style guide, create SVG components
  6. Implement animations → code scenes in Remotion
  7. Quality assurance → auto-check, auto-fix, auto-start preview

⚠️ Context Recovery Protocol

Every conversation turn may follow a context loss (compaction, new session). Before doing ANY work:

  1. Check if remotion_video/PROGRESS.md exists

- If YES → Read it completely to determine current phase and last completed step - If NO → This is a new project, proceed to Phase 1

  1. Read supporting files referenced in PROGRESS.md (only if that phase is marked complete):

- remotion_video/script.md (if Phase 1.5+ completed) - remotion_video/storyboard.md (if Phase 2+ completed) - src/<Composition>/constants.ts (if Phase 3+ completed — contains COLORS, SCENES, NARRATION, AUDIO_SEGMENTS)

  1. Verify files listed in "Files Created" section actually exist on disk
  2. Resume from the first unchecked item in the current phase
Skipping this protocol causes repeated work or file corruption. Always run it first.

Progress Tracking — Mandatory Protocol

⚠️ This protocol is NON-NEGOTIABLE. Skipping updates causes context loss and repeated work.

Maintain remotion_video/PROGRESS.md using progress-template.md. Create it at Phase 1 start. Log decisions in the Decisions table and add every created file to "Files Created".

Checkpoint Rule

Every time you complete a checkbox item in PROGRESS.md, you MUST immediately:

  1. Mark the item [x] and add brief notes
  2. Update the "Current State" section (Current Phase + Current Step)
  3. Then — and only then — proceed to the next item

Do NOT batch multiple items. One item done → one update → next item.

Phase Transition Gate

Before starting any new Phase, you MUST:

  1. Read PROGRESS.md and verify ALL items in the previous phase are [x]
  2. Update "Current Phase" to the new phase
  3. If any previous item is unchecked, complete it first

Workflow

Phase 1: Requirements Gathering

⚠️ First: Complete the Prerequisites section (install remotion-best-practices skill and read it). Then create remotion_video/PROGRESS.md from progress-template.md and fill in Project Info.

Before starting, confirm these essential details with the user:

  • Topic: What concept/subject to explain?
  • Audience: Who is watching? (children/adults, beginners/experts)
  • Language: Chinese/English/other?
  • Duration: Short (1-3min), Medium (3-5min), or Long (5-10min)?
  • Key points: What must the viewer learn?

For detailed question templates, see requirements-guide.md.

Phase 1.5: Script Writing

⚠️ Checkpoint Rule active: After completing EACH checkbox item for this phase, immediately update PROGRESS.md. Do not batch updates.

Write a complete narrative script before designing the storyboard. This phase focuses purely on storytelling — what to say and how to say it well — without worrying about visual specs, frame numbers, or animation parameters.

IMPORTANT: Write the FULL narration text — every word that will be spoken. Do NOT write summaries, bullet points, or placeholders. The script is the final spoken content.

The script must include:

  1. Core message — one-line summary, learning objectives
  2. Narrative strategy — apply techniques from script-and-narration.md:

- Entry angle (question / scenario / challenge / story) - Core metaphor that runs through the entire video - Knowledge scaffolding order (what depends on what) - Emotional curve (curiosity → understanding → wonder → satisfaction)

  1. Full narration text — complete word-for-word script for every chapter:

- Include emphasis markers (bold for stress, *italic* for softer tone) - Mark pauses with [.] (short), [..] (medium), [...] (long), [PAUSE] or [BEAT] (dramatic) — see script-and-narration.md Part 3 for duration semantics - Add visual intents after each chapter (1-2 sentences describing what viewers should see — enough for Phase 2 to design scenes, but no animation specs)

  1. Pacing notes — where to speed up, slow down, and pause
  2. Self-review — run through the checklist in script-and-narration.md before presenting to user

Quality gate: Present the complete script to the user for approval. Do NOT proceed to Phase 2 until the user explicitly approves the narrative.

Why script first:

  • Separates "what to tell" from "how to show" — two different creative activities
  • LLM produces better narratives when not simultaneously calculating frame ranges
  • Pure text is cheap to iterate; storyboard with animation specs is expensive to revise
  • Users can easily review "is the story good?" without wading through technical details

Output: Save the approved script as remotion_video/script.md

See script-and-narration.md for video structure templates, narrative techniques, writing techniques, and TTS notes.

Phase 2: Storyboard Design

⚠️ Checkpoint Rule active: After completing EACH checkbox item for this phase, immediately update PROGRESS.md. Do not batch updates.

Convert the approved script into a production-ready storyboard. The script provides what to say; the storyboard defines how to show it.

Input: Completed script (approved in Phase 1.5)

Tasks:

  1. Break script chapters into visual scenes (5-15 scenes)
  2. Assign narration text from the script to each scene
  3. Design visual layers for each scene (background / midground / foreground / UI)
  4. Add frame-level animation specifications (spring, easing, timing)
  5. Define visual-narration sync points
  6. Plan the asset inventory (SVG components, colors, typography)

The cognitive load is much lower than creating everything from scratch — the narrative is already decided, so you only need to focus on visual execution.

Output: Save the completed storyboard as remotion_video/storyboard.md for design traceability and iteration reference.

See storyboard-template.md for templates. See script-and-narration.md Part 4 for subtitle formatting and TTS notes.

Phase 3: Visual Design

⚠️ Checkpoint Rule active: After completing EACH checkbox item for this phase, immediately update PROGRESS.md. Do not batch updates.

Apply the Kurzgesagt/回形针 style. Concrete steps:

  1. Choose color palette — Select from design-tokens.ts Section 1 or create a custom palette matching the topic
  2. Create constants.ts — Define COLORS, TYPOGRAPHY, SCENES, NARRATION, and estimated AUDIO_SEGMENTS following design-tokens.ts Section 3
  3. Set up fonts — Use Remotion's loadFont() from @remotion/google-fonts (see design-tokens.ts Section 2 for reference)
  4. Create shared SVG components — Build reusable visual elements (icons, illustrations, decorative elements) following svg-components.md
  5. Design scene layouts — Plan visual layers (background / midground / foreground / UI) per scene following visual-principles.md

Style principles:

  • Flat design with subtle gradients
  • Bold, saturated color palette
  • Geometric shapes with rounded corners (rx/ry)
  • Clean sans-serif typography

See style-guide.md for complete visual standards. See visual-principles.md for composition and layout.

Phase 4: Animation Production

⚠️ Checkpoint Rule active: After completing EACH checkbox item for this phase, immediately update PROGRESS.md. Do not batch updates. Log key file paths in "Files Created".

Implement scenes using Remotion:

  1. Create SVG components for visual elements
  2. Use useCurrentFrame() for all animations
  3. Apply appropriate easing (spring for natural motion)
  4. Add scene transitions

Subtitle & narration rules (critical for Phase 4.5 compatibility):

  • All narration text must be stored in the NARRATION object in constants.ts — never hardcode text directly in scene TSX files
  • Create an estimated AUDIO_SEGMENTS in constants.ts with approximate timing. Phase 4.5 will overwrite it with real audio-based timing
  • Subtitle components must reference AUDIO_SEGMENTS.sceneKey — never use inline segment arrays with hardcoded frame numbers
  • AUDIO_SEGMENTS 中的 startFrame/endFrame 必须使用场景本地帧号(每个场景从 SCENE_PAD=15 开始),不是全局帧号。因为 AudioLayer 和 SubtitleSequence 都在场景的 <Sequence> 内部运行,useCurrentFrame() 返回的是本地帧号。如果使用全局帧号,后续场景的字幕会延迟或完全不显示
  • This ensures rebuild-timeline.ts --write in Phase 4.5 can update timing without modifying any scene files

Visual-narration alignment rules (prevents animation-subtitle desync):

  • 每个与旁白内容对应的视觉元素(图标出现、箭头展开、图表绘制等),其 startFrame 必须从 AUDIO_SEGMENTS 对应段的 startFrame 派生,不能凭"视觉节奏"硬编码
  • 正确模式:const liftArrowStart = AUDIO_SEGMENTS.forces[0].startFrame;(升力箭头在旁白说"升力"时出现)
  • 错误模式:const liftArrowStart = 30;(凭感觉写的帧数,和旁白无关)
  • 纯装饰性动画(背景粒子、环境氛围)不受此约束,可以自由定时
  • Phase 4 使用估算 AUDIO_SEGMENTS;Phase 4.5 rebuild-timeline 更新真实时间后,因为代码引用的是变量而非硬编码数字,视觉动画会自动同步
  • 参考 animation-guide.md "Narration-Synced Animation" 章节的实现模式

Element sizing rules (prevents "Thumbnail Syndrome" — tiny elements on a large canvas):

  • 参考 visual-principles.md "Content Area Utilization" 确保内容填充画布(核心内容 ≥ 安全区 60%)
  • 组件默认尺寸参考 svg-components.md(图标 ≥96px,流程节点 ≥160px 高,标签 ≥40px)
  • 图表/流程图应占据内容区域大部分面积,避免在 1920×1080 画布上缩成缩略图

Background rules (prevents transparent/checkerboard frames during transitions):

  • The main composition must have a persistent <AbsoluteFill> background layer (using COLORS.background) that sits behind all scenes and never participates in transitions
  • Each scene component must also have its own solid background as the first child element
  • During fade() transitions, both scenes have reduced opacity — without a persistent background, transparent frames appear as a checkerboard pattern in preview and black in renders
  • See animation-guide.md "Preventing Transparent Frames" for the implementation pattern

Visual richness rules (prevents PPT-like output):

  • 每个场景必须有至少一个 非文字的视觉主体元素(SVG 插画、图表、动画图形等)。纯文字标签 + 方块不是合格的视觉内容
  • 流程图/因果链必须用 图标或插画 配合文字,不能只用纯色方块装文字。参考 svg-components.md "Illustrated Flow Node" 模式
  • 每个场景应有 环境氛围层(粒子、光晕、纹理等),参考 style-guide.md Ambient Effects 章节
  • SVG 插画应体现 Kurzgesagt 风格:圆角几何形状(rx/ry)、扁平化但有层次(多 path 叠加)、柔和渐变(linearGradient/radialGradient)、适当描边
  • 参考 visual-principles.md "Show, Don't Tell" 原则:能用图示表达的概念不要用文字方块
  • 参考 scene-template.tsx 中 ForceDiagramScene 的 SVG 飞机插画作为具象插画的最低质量标准

Color rules (critical for Phase 5 style-scan compliance):

  • All colors must be referenced via the COLORS object from constants.ts (e.g., COLORS.accent.rose) — never write hex values directly in TSX files
  • The only exception is rgba() for opacity variations (e.g., rgba(0, 0, 0, 0.7) for subtitle backgrounds)
  • This prevents the common issue where style-scan reports dozens of "color not in approved palette" warnings

See design-tokens.ts Section 3 for the complete constants.ts structure (COLORS, SCENES, NARRATION, AUDIO_SEGMENTS, font loading). See svg-components.md for component patterns. See animation-guide.md for timing and easing.

Phase 4.5: Audio Generation

⚠️ Checkpoint Rule active: After completing EACH checkbox item for this phase, immediately update PROGRESS.md. Do not batch updates.

完成动画编码后,生成视频音频并同步时间线:

  1. 生成 TTS 音频 — 用 generate-tts.ts 从 NARRATION 提取文本生成 mp3
  2. 重建时间线 — 用 rebuild-timeline.ts --write 根据实际音频时长更新 constants.ts
  3. 调整动画关键帧 — 按 newDuration / oldDuration 比例缩放
  4. 添加背景音乐 — 免版税 BGM 到 public/audio/bgm.mp3
  5. 创建 AudioLayer — 播放旁白 + BGM 的组件
  6. 集成并验证 — AudioLayer 必须在 TransitionSeries 外部

详细步骤、命令参数、AudioLayer 模板见 audio-guide.md。

Phase 5: Quality Assurance

⚠️ Checkpoint Rule active: After completing EACH checkbox item for this phase, immediately update PROGRESS.md. Do not batch updates.

完成编码后,执行自动质量检查流程:

  1. 代码扫描 → 检查样式合规性(字号、颜色、安全区域等)
  2. 截图审查 → 渲染关键帧,视觉检查
  3. 自动修复 → 根据检查报告修复问题,循环直到通过
  4. 启动项目 → 所有检查通过后,自动启动 Remotion 预览

详细检查步骤和规则见 quality-checklist.md。

Phase 6: Final Export

After preview looks correct, render the final video:

cd remotion_video
npx remotion render src/index.ts <CompositionName> out/video.mp4

Options:

  • --codec h264 (default) or --codec h265 for smaller file size
  • --quality 80 to 100 (default: 80)
  • --scale 1 (1080p) — use --scale 2 for 4K if source assets support it
  • Add --log verbose if debugging render issues

The output file will be at remotion_video/out/video.mp4.

Video Structure

Standard educational video structure:

1. Hook (5-10s)      - Attention-grabbing question or statement
2. Intro (10-20s)    - Topic introduction
3. Content (main)    - Core explanation, broken into segments
4. Summary (10-20s)  - Key takeaways
5. Outro (5-10s)     - Call to action or closing

See script-and-narration.md Part 1 for detailed structure templates.

Key Principles

Content Clarity

  • One concept per scene
  • Build complexity gradually
  • Use visual metaphors for abstract ideas

Visual Simplicity

  • Minimal elements on screen
  • Clear visual hierarchy
  • Consistent style throughout

Animation Purpose

  • Every animation serves understanding
  • Avoid decorative motion
  • Sync with narration pace

Reference Files

FileWhen to Use
requirements-guide.mdStarting a new video project (Phase 1)
script-and-narration.mdVideo structure, script writing, narration, subtitle/TTS (Phase 1.5 + 2)
storyboard-template.mdConverting script into visual scenes (Phase 2)
style-guide.mdDesigning visual elements (Phase 3 + 4)
visual-principles.mdLayout and composition decisions (Phase 3 + 4)
animation-guide.mdImplementing animations (Phase 4)
svg-components.mdCreating reusable components (Phase 4)
audio-guide.mdTTS 生成、时间线重建、AudioLayer 集成详细步骤 (Phase 4.5)
quality-checklist.mdQuality assurance + style check rules (Phase 5)
design-tokens.tsColor palettes, typography presets, constants.ts template (Phase 3)
progress-template.md执行进度跟踪模板 (全流程)
scene-template.tsx场景组件模板 (Phase 4)
subtitle-sequence-template.tsx字幕序列组件模板 (Phase 4)
common-icons.tsx通用 SVG 图标组件 (Phase 4)
generate-tts.ts字幕提取 + TTS 音频生成脚本 (Phase 4.5)
rebuild-timeline.ts音频时长测量 + 时间线重建脚本 (Phase 4.5)
style-scan.ts代码样式扫描脚本 (Phase 5)
render-keyframes.ts关键帧批量截图脚本 (Phase 5)
shared.ts脚本共享函数(内部依赖,无需直接调用)

平台分布

Codex

40.24%
按下载量换算945

Claude

29.99%
按下载量换算704

Cursor

18.2%
按下载量换算427

Gemini CLI

8.68%
按下载量换算204

安全审计

Gen Agent Trust Hub

未通过

Socket

通过

Snyk

可疑

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills