home / skills / team-attention / plugins-for-claude-natives / youtube-digest

This skill analyzes a YouTube URL to extract transcript, generate a concise summary and translation, and creates a 9-question quiz to test comprehension.

npx playbooks add skill team-attention/plugins-for-claude-natives --skill youtube-digest

Review the files below or copy the command above to add this skill to your agents.

Files (5)
SKILL.md
3.1 KB
---
name: youtube-digest
description: This skill should be used when the user asks to "유튜브 정리", "영상 요약", "transcript 번역", "YouTube digest", "영상 퀴즈", or provides a YouTube URL for analysis. Extracts transcript, generates summary/insights/Korean translation, and tests comprehension with 9 quiz questions across 3 difficulty levels. Optional Deep Research for web-based follow-up.
---

# YouTube Digest

YouTube 영상 분석 → 요약/인사이트/번역 문서 생성 → 퀴즈 테스트.

## 워크플로우

### 1. 메타데이터 수집

```bash
scripts/extract_metadata.sh "<URL>"
```

추출: title, description, channel, upload_date, duration, tags

### 2. Transcript 추출

```bash
scripts/extract_transcript.sh "<URL>" [output_dir]
```

우선순위: 수동 자막(ko→en) > 자동 생성 자막(ko→en)

### 3. 맥락 파악 (WebSearch)

웹 검색으로 고유명사 정확한 표기 수집:
- `"{영상 제목}" {채널명} summary`
- `"{발표자명}" {주제 키워드}`

### 4. Transcript 교정

자동 자막의 고유명사 오인식을 웹 검색 결과로 대체:
- Kora → Cora, cloud code → Claude Code, every → Every.to

### 5. 문서 생성

```markdown
---
title: {영상 제목}
url: {YouTube URL}
channel: {채널명}
date: {업로드 날짜}
duration: {영상 길이}
processed_at: {처리 일시}
---

# {영상 제목}

## 요약
{3-5문장 요약 + 주요 포인트 3개}

## 인사이트
### 핵심 아이디어
### 적용 가능한 점

## 전체 스크립트 (한글 번역)
[00:00] ...
```

### 6. 파일 저장

위치: `research/readings/youtube/{YYYY-MM-DD}-{sanitized-title}.md`

### 7. 학습 퀴즈

3단계 × 3문제 = 총 9문제. AskUserQuestion으로 각 단계 3문제 동시 출제.

| 단계 | 난이도 | 출제 기준 |
|------|--------|----------|
| 1 | 기본 | 핵심 인사이트, 주요 개념 |
| 2 | 중급 | 인사이트 + 세부 내용 연결 |
| 3 | 심화 | 세부 내용, 적용/분석 |

문제 유형 상세: `references/quiz-patterns.md`

#### 결과 처리

틀린 문제에 대해 정답과 해설 제공 후, 문서 끝에 퀴즈 결과 추가:

```markdown
## 퀴즈 결과

총점: 7/9 (78%) | 1단계 3/3 ✅ | 2단계 2/3 | 3단계 2/3

### 오답 노트

**Q5**: {질문}
- 선택: B → 정답: C
- {1-2문장 해설}
```

### 8. 후속 선택

퀴즈 완료 후 AskUserQuestion:
- **한 번 더 퀴즈**: 다른 문제로 재테스트
- **Deep Research**: 웹 심층 조사 (`references/deep-research.md` 참조)
- **종료**: 마무리

## 참고사항

### 자막 언어 우선순위
1. 한국어 수동 → 2. 영어 수동 → 3. 한국어 자동 → 4. 영어 자동

### 불완전한 자막 처리
- 고유명사 오인식: 4단계에서 일괄 대체
- 이해 불가 부분: `[불명확]` 표시

### yt-dlp 옵션
- `--list-subs`: 자막 목록 확인
- `--cookies-from-browser chrome`: 로그인 필요 시

## 리소스

- `scripts/extract_metadata.sh` - 메타데이터 추출
- `scripts/extract_transcript.sh` - 자막 추출
- `references/quiz-patterns.md` - 퀴즈 문제 유형 상세
- `references/deep-research.md` - Deep Research 워크플로우

Overview

This skill extracts and analyzes YouTube videos to produce a consolidated digest: cleaned transcript, concise summary, insights, Korean translation, and a 9-question comprehension quiz across three difficulty levels. It supports optional web-based Deep Research to verify names and clarify ambiguous transcript segments. Outputs are saved as a structured markdown document for easy review and follow-up.

How this skill works

The skill fetches video metadata and the best-available transcript (preferring manual subtitles over auto-generated ones), runs web searches to confirm proper nouns and context, and corrects transcript errors. It generates a 3–5 sentence summary with three key points, insight sections, a full Korean translation of the transcript, and a 3x3 quiz (easy/medium/hard). It can then offer further actions: retake quiz, run Deep Research, or finish.

When to use it

  • Requesting a YouTube digest, Korean translation of a transcript, or video summary
  • Providing a YouTube URL for analysis
  • Preparing study material or training content from a video
  • Assessing comprehension with automated quizzes
  • Verifying names/terms when transcript quality is poor

Best practices

  • Provide the YouTube URL and preferred language for translation (default: Korean)
  • If available, allow use of cookies or authenticated access for private/subscribed captions
  • Choose Deep Research when the video contains many proper nouns or technical terms
  • Review the corrected transcript for any remaining [unclear] markers before finalizing
  • Use the saved markdown file location for long-term study and versioning

Example use cases

  • Create a study guide and 9-question quiz for a technical talk to use in a team learning session
  • Translate an English lecture to Korean and get a concise summary plus actionable insights
  • Generate corrected transcript and metadata for documentation or knowledge base ingestion
  • Run Deep Research on a video with many unfamiliar names to ensure accurate terminology
  • Use the quiz output to test trainees and append results and explanations to the digest document

FAQ

What subtitle languages are preferred?

Priority: manual Korean → manual English → auto Korean → auto English. The skill picks the best available in that order.

How are quiz difficulty levels defined?

Three levels: Level 1 (basic) focuses on core ideas; Level 2 (intermediate) links insights to details; Level 3 (advanced) tests deep understanding and application/analysis with real-world scenarios.

What happens if the transcript is poor quality?

The system marks unclear segments as [unclear], runs web searches to disambiguate proper nouns, and replaces likely misrecognitions before generating outputs.