将YouTube视频/播放列表转换为 SKILL.md 文件,让Claude Code能够从视频内容中学习新的技能和工作流程
概览
YouTube上有数百万个编程教程,但这些知识是: 1. — AI 代理无法观看视频 2.— 一个10分钟的视频观看时间为10分钟 3.— API 变动,库弃用,最佳实践演变 4.— 叙述格式(“那么我们要做什么......”)≠可执行指令 人工智能代理技能扩展了代理的能力。然而,创造有效的技能需要理解: - 代理如何发现技能(通过描述匹配)- 叙述性内容与命令性内容的区别- 上述技术的现状 没有这些知识,生成的技能通常如下:- 未激活(描述不准确)- 无法使用(指令不完整)- 有害(信息过时) 该技能自动化了转化流程: 使用字幕API(带Whisper备用)从YouTube抓取文字稿: : JSON 包含元数据(标题、频道、日期)和完整文字稿。
概述(AI 翻译)
将YouTube教程转化为真正有效的AI代理技能。
问题
视频知识被困住了
YouTube上有数百万个编程教程,但这些知识是:
- AI助手无法访问——AI代理无法观看视频
- 提取耗时 — 一个10分钟的视频观看时间为10分钟
- 常常过时——API变更,库弃用,最佳实践演进
- 不可操作 — 叙述格式(“那么我们要做什么…”)≠可执行指令
技能质量问题
人工智能代理技能扩展了代理的能力。然而,创造有效的技能需要理解:
- 智能体如何发现技能(通过描述匹配)
- 叙述性内容与命令性内容的区别
- 上述技术的现状
没有这些知识,生成的技能通常会:
- 从未激活(描述不准确)
- 不可用(指令不完整)
- 有害(过时信息)
解决方案
该技能自动化了转化流程:
1.提取物
使用字幕API(带Whisper备用)从YouTube抓取文字稿:
本页由 AI 翻译,技术细节请以原文为准。
README
以下为原文 README 节选,完整内容请查看原文。
YouTube to Skill
Transform YouTube tutorials into AI agent skills that actually work.
The Problem
Video Knowledge is Trapped
YouTube contains millions of programming tutorials, but this knowledge is:
- Inaccessible to AI assistants — AI agents cannot watch videos
- Time-consuming to extract — A 10-minute video takes 10 minutes to watch
- Often outdated — APIs change, libraries deprecate, best practices evolve
- Not actionable — Narrative format (“so what we’re gonna do…”) ≠ executable instructions
The Skill Quality Problem
AI agent skills extend an agent’s capabilities. However, creating effective skills requires understanding:
- How agents discover skills (via description matching)
- The difference between narrative and imperative content
- Current state of mentioned technologies
Without this knowledge, generated skills are often:
- Never activated (poor descriptions)
- Unusable (incomplete instructions)
- Harmful (outdated information)
The Solution
This skill automates the transformation pipeline:
┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ EXTRACT │────▶│ IDENTIFY │────▶│ VERIFY │────▶│ TRANSFORM │────▶│ GENERATE │
│ transcript │ │technologies │ │ up-to-date │ │ content │ │ skill │
└─────────────┘ └─────────────┘ └─────────────┘ └─────────────┘ └─────────────┘
1. Extract
Pulls transcripts from YouTube using captions API (with Whisper fallback):
python scripts/extract_youtube.py "https://youtube.com/watch?v=..."
Output: JSON with metadata (title, channel, date) and full transcript.
2. Identify
Scans transcript for technologies:
- Languages and versions
- Frameworks and libraries
- APIs and services
- CLI tools
3. Verify
Videos get outdated. Skills must not.
Uses available tools to verify current state:
| Source | Tool | Use Case |
|---|---|---|
| Library docs | Context7 MCP[^2] | React, Node.js, Python packages |
| General search | WebSearch | Tools, services, APIs |
| Specific docs | WebFetch | Official documentation URLs |
When outdated content is detected, the user is asked how to proceed:
- Update to current version
- Keep original with warnings
- Skip the section
4. Transform
Converts narrative video content to actionable skill instructions:
| Video Content | Skill Content |
|---|---|
| “So basically what we’re doing…” | (removed) |
| “You want to run this command…” | bash\ncommand\n |
| “The important thing here is…” | Prerequisites section |
| “If you get this error…” | Troubleshooting section |
5. Generate
Produces a SKILL.md following the standard skill format:
---
name: lowercase-with-hyphens
description: Action verb + what + when to use (max 1024 chars)
---
File Structure
推荐工具
换一个关键词,或者移除筛选条件。
安装
npx skillfish add yfe404/ai-anything