| name | youtube-summary |
| description | Summarize YouTube videos from a URL, youtu.be link, Shorts link, embed URL, or raw video ID. Use when I share a YouTube link and ask "요약", "정리", "transcript", "자막", "강의 노트", "영상 내용", "이거 봐줘", or want an Obsidian-ready Korean note with timestamps, key claims, quotes, and follow-up questions. Also use when I ask for "lecture-summarization", "강의 요약", "긴 글로 재구성", or a comprehensive lecture-style article. Fetch the transcript first; do not invent content when captions are unavailable. |
YouTube Summary
Turn a YouTube video into a concise Korean working note grounded in the transcript.
Procedure
-
Extract the transcript first. Prefer the bundled helper:
uv run skills/productivity/youtube-summary/scripts/fetch_transcript.py "<youtube-url>" --language ko,en --timestamps
If the user asks only for raw transcript, add --text-only. If the target language is
clear from the request, put it first in the language list.
-
Validate the transcript before summarizing: non-empty, plausible language, timestamps
present. If the requested language is unavailable, retry with --language en,ko, then
with no language flag.
-
For long transcripts, chunk by timestamps. Summarize each chunk, then merge; preserve
the timestamp range for any claim that depends on a specific section.
-
Choose the output mode:
- Use Default output unless the user asks for a different format.
- Use lecture-summarization when the user explicitly says
lecture-summarization
or asks for a comprehensive lecture/article reconstruction.
-
Output Korean by default unless the user asks otherwise. Keep it Obsidian-ready and
skimmable.
-
If captions are disabled, unavailable, private, region-blocked, or the helper fails,
say that directly and do not summarize from the title/thumbnail alone. Offer the next
useful fallback: ask for pasted transcript, audio file, or another accessible source.
Default output
Use this shape unless the user asks for a different format:
# <video title or concise topic>
## 한 줄 요지
- <one sentence>
## 핵심 정리
- <claim> (`MM:SS`)
- <claim> (`MM:SS`)
- <claim> (`MM:SS`)
## 챕터
- `00:00` <section title> — <one-sentence summary>
## 인용 후보
- `MM:SS` <short paraphrased quote or exact quote if brief>
## 내 작업에 연결
- <why this matters / how I might use it>
## 후속 질문
- <question>
Option: lecture-summarization
Use this mode when the user asks for lecture-summarization, "강의 요약",
"긴 글로 재구성", or a comprehensive article rather than a compact working note.
This mode reconstructs the transcript into a long-form reference article. It should be
expansive, but still grounded in the transcript. Do not add external facts unless the user
explicitly asks for research.
Before writing, infer:
video_minutes: video length from metadata if available; otherwise estimate from the
final transcript timestamp.
target_read_minutes: max(5, video_minutes // 4) for videos longer than 15 minutes,
otherwise 5.
output_language: Korean by default unless requested otherwise.
Use this instruction frame:
You are an expert editor at a prestigious knowledge publication like Harvard Business
Review or MIT Technology Review.
Your task is to reconstruct a video transcript into a definitive, long-form article.
## Critical Context
- Video duration: <video_minutes> minutes
- Target reading time: ~<target_read_minutes> minutes
- Output language: <output_language>
- Critical rule: do not make a short abstract. Expand and elaborate. For long videos,
each minute of video should translate into roughly 150-200 words of article text when
the transcript supports that level of detail.
## Mission
Transform the raw transcript into authoritative reference material. A 5-minute read for a
long lecture is a failure; provide a deep-dive experience that justifies the target
reading time.
## Expansion Guidelines
1. Detail restoration: include every substantive anecdote, case study, supporting
argument, and story arc from the speaker.
2. Structural depth: each major `##` section should contain several dense paragraphs when
the transcript has enough material.
3. No omission: do not skip minor points when they carry nuance, assumptions, constraints,
trade-offs, or examples.
4. Technical elaboration: when a technical term appears, explain its context, importance,
and relationship to nearby concepts in the lecture.
## Content Requirements
- All major concepts and granular sub-points
- All examples, analogies, and step-by-step methodologies
- All key statistics and data points
- Comparisons, pros/cons, and trade-offs
- Common misconceptions and the speaker's rebuttals
## Writing Standards
- Write entirely in <output_language>.
- Tone: formal, analytical, and sophisticated.
- Structure:
# <compelling scholarly headline>
*~<target_read_minutes>분 읽기*
<Hook paragraph of 150+ words that sets the stage and explains why the topic matters now.>
## <Major Topic 1>
<Extensive explanation grounded in the transcript.>
## <Major Topic 2>
<Extensive explanation grounded in the transcript.>
<Add as many sections as needed to cover the duration proportionally.>
## Key Takeaways (Deep Analysis)
- **<Concept 1>**: <3-4 sentence analysis, not just a bullet summary.>
## 핵심 용어 및 개념 상세 정리
| 용어 | 맥락 및 심층 정의 |
|------|----------------|
| <Term 1> | <How this term was used in the lecture and why it matters.> |
For very long transcripts, produce chunk-level reconstructions first, then merge them into
one article. Preserve timestamps only when they help audit a specific claim; do not turn
this mode into a timestamped outline.
Guardrails
- Do not present guesses as transcript-grounded facts.
- Keep direct quotes short and timestamped; paraphrase by default.
- Do not include sponsor reads, intros, or repeated calls to action unless they matter.
- Preserve uncertainty: if auto-captions are noisy, say so and mark weak claims.
- For Korean notes, use direct, technical prose with no filler.