YouTube Intelligence Extractor — YouTube Intelligence 提取器

YouTube Intelligence 提取器 Overview

This 技能 turns YouTube video transcripts into structured, actionable intelligence across four domAIns that matter to Chibitek Labs: productivity, AI prompting, 平台/系统s engineering, and creative building. Instead of passive consumption, every video becomes a reusable as设置 — 提取ed todos, prompts you can 运行 immediately, 框架s you can 应用ly, and insights m应用ed to your current work.

After presenting the 报告, always save it as a markdown file and present it to the user. This is standing behavior — do not wAIt to be asked.

Quick Reference Situation 应用roach User gives a YouTube URL Try Method A (Wave Tube fetch) → fallback to Method B (python) → fallback to Method C (manual paste) User pastes raw transcript Skip fetch, go strAIght to 提取ion 流水线 Python fetch is IP-blocked Use Wave Tube 网页 fetch — it works from server 环境s Video has no captions Attempt auto-生成d captions; if unavAIlable, 通知 user User wants one specific domAIn 运行 full 流水线, present only the 请求ed domAIn User wants everything 运行 full 流水线, present all four intelligence 报告s Very long video (1hr+) Chunk transcript, process in sections, merge 输出s 报告 is complete Always save as markdown file and present_files — do not ask Step 1: Fetch the Transcript Method A — Wave Tube (preferred — works from server/cloud 环境s)

Wave Tube mirrors YouTube transcripts and is not IP-blocked. Always try this first.

URL pattern: https://tube.wave.co/[video-slug]-[VIDEO_ID]

To find the slug: 搜索 "[VIDEO_ID] youtube" — the Wave Tube 结果 will 应用ear. Or 网页_搜索 for the video title + video ID; Wave Tube typically 应用ears in 结果s.

Use the 网页_fetch 工具 on the Wave Tube URL. The full transcript 应用ears in the page body under "## Transcript". Also captures: title, channel, views, duration, description, and timestamps in the same fetch.

Why Wave Tube first: youtube-transcript-API fAIls with IpBlocked when Claude 运行s on cloud infrastructure (AWS/GCP/Azure). Wave Tube is a reliable public mirror that bypasses this entirely and also provides video metadata in the same fetch.

Method B — Python (works locally on macOS, fAIls in server 环境s) pip 安装 youtube-transcript-API --break-系统-packages -q

from youtube_transcript_API 导入 YouTubeTranscriptAPI 导入 re

def 提取_video_id(url): patterns = [ r'(?:v=|youtu\.be/)([a-zA-Z0-9_-]{11})', r'(?:embed/)([a-zA-Z0-9_-]{11})', ] for p in patterns: m = re.搜索(p, url) if m: return m.group(1) return url # assume it's already a video ID

# 导入ANT: Use instance method .fetch(), NOT .获取_transcript() # The API changed in v1.0+ — always instantiate first video_id = 提取_video_id("VIDEO_URL_HERE") transcript = YouTubeTranscriptAPI().fetch(video_id) full_text = " ".join([entry.text for entry in transcript]) print(full_text)

Common mistake: YouTubeTranscriptAPI.获取_transcript() no longer exists in v1.0+. Always use YouTubeTranscriptAPI().fetch(video_id) — instantiated, not class method.

Method C — Manual paste (always works, zero dependencies)

If 机器人h automated methods fAIl, ask the user to:

Go to the YouTube video 命令行工具ck the ... menu → Show transcript Copy all the transcript text Paste it directly into the chat

Then proceed with the 提取ion 流水线 on the pasted text.

Step 2: 运行 the 提取ion 流水线

Once you have the transcript text, 提取 across all four domAIns. Each domAIn is independent — process them 清理ly and present them in sequence.

DomAIn 1 — Todos & Action Items

提取 concrete, doable actions mentioned or implied in the video.

输出 格式化:

✅ Todos & Action Items

• [ ] [Specific action] — [上下文, 工具, or condition if mentioned]

Rules:

Every item must be specific enough to act on without watching the video Include 工具 names, URLs, timeframes where mentioned Do NOT include vague advice ("learn more about X") — skip it or make it concrete DomAIn 2 — Advice & 框架s

提取 principles, mental 模型s, 框架s, and strategic advice that 应用ly repeatedly — not just one-time instructions.

输出 格式化:

💡 Advice & 框架s

[框架 Name]

DomAIn 3 — AI Prompts (提取able & Ready to Use)

提取 any AI prompts, prompt strategies, or prompt patterns mentioned, shown, or implied. Reconstruct them as 运行nable prompts — not descriptions of prompts.

输出 格式化:

🤖 AI Prompts & Techniques

[Technique Name]

[Full, ready-to-运行 prompt text in a fenced code block]

Notes: [模型, 上下文, or usage tips]

Rules:

If a prompt is partially shown, reconstruct the most likely complete version 格式化 prompts in fenced code blocks so they are easy to copy Never just describe a prompt — always write the actual text DomAIn 4 — 平台, Engineerin