Youtube Summarizer

Download YouTube video transcripts and generate structured summaries with key viewpoints, notable quotes, and topic analysis. Use when the user shares a YouTube URL and asks to summarize the video, extract transcripts, or get content insights. Supports any YouTube URL format (watch, youtu.be, shorts, embed, live).

Download YouTube video transcripts/subtitles and generate structured content summaries with key viewpoints, notable quotes, and topic analysis.

Setup

Install dependencies (skip if already installed):

bash

pip3 install youtube-transcript-api -q
# yt-dlp is needed for video metadata (title, channel, duration)
# Install if not present: brew install yt-dlp  OR  pip3 install yt-dlp

Workflow

Step 1: Fetch Transcript

SKILL_DIR is the directory containing this SKILL.md file.

bash

# JSON output with metadata
python3 SKILL_DIR/scripts/fetch_transcript.py "https://youtube.com/watch?v=VIDEO_ID"

# Plain text with timestamps (recommended for summarization)
python3 SKILL_DIR/scripts/fetch_transcript.py "URL" --text-only --timestamps

# Specific language with fallback chain
python3 SKILL_DIR/scripts/fetch_transcript.py "URL" --language en,zh --text-only --timestamps

The script accepts any standard YouTube URL format: watch?v=, youtu.be/, shorts/, embed/, live/, or a raw 11-character video ID.

Step 2: Fetch Video Metadata

Get title, channel, duration, and upload date for context:

bash

yt-dlp --skip-download --print "%(title)s|%(channel)s|%(duration_string)s|%(upload_date)s" "URL"

Step 3: Read Full Transcript

For long videos (>30 min), the transcript output may be large. Read in sections:

bash

# First half
python3 SKILL_DIR/scripts/fetch_transcript.py "URL" --text-only --timestamps | head -500
# Second half
python3 SKILL_DIR/scripts/fetch_transcript.py "URL" --text-only --timestamps | tail -400

Step 4: Generate Summary

Analyze the transcript and produce a structured summary. ALL content must be in the chosen language — including section labels, table headers, and metadata fields. Only the template structure (emojis, table format, blockquote format) stays fixed. The actual words are all localized.

Template structure (translate every label into the output language):

text

## [localized: "视频总结" / "Video Summary"]

**[localized label]:** (title)
**[localized label]:** (channel)
**[localized label]:** (guest, if identifiable)
**[localized label]:** (duration) | **[localized label]:** (date)

---

### [localized: one-line summary]
(1-2 sentence overview)

---

### [localized: "核心观点" / "Key Viewpoints"]

| [localized: 话题] | [localized: 观点] |
|-------|-----------|
| ... | ... |

---

### [localized: "金句" / "Notable Quotes"]
> "Original language quote" — Speaker

---

### [localized: "其他亮点" / "Other Highlights"]
- ...

---

### [localized: "注意/偏见" / "Notes / Bias"]
- ...

Output Language

The summary language is determined by this priority:

1. User explicitly specifies (e.g. "用中文总结", "summarize in Japanese", "用英语总结") → use that language
2. User does not specify → default to 中文 (Chinese) for ALL content
3. Quotes → always keep in the original spoken language (do not translate quotes)
4. Section labels, table headers, metadata fields → translate into the output language (中文 by default)
5. Only structural format (tables, blockquotes, bold) stays fixed — no emojis

Error Handling

Error	Solution
youtube-transcript-api not installed	Run pip3 install youtube-transcript-api
Transcripts are disabled	Tell user subtitles are disabled for this video
No transcript found	Retry without --language flag, then note actual language
Empty output	Video may not have auto-generated subtitles; inform user
yt-dlp: command not found	Install with pip3 install yt-dlp or brew install yt-dlp

Reference Example

See references/example-output.md for a real summary (GLM-5.2 interview video, Chinese output). Use it as the quality bar for structure, conciseness, and depth.

Tips

• Speaker identification: For interview/podcast videos, identify speakers from the intro and label viewpoints accordingly
• Sponsor segments: Skip ad reads (common in tech videos) — note them in ⚠️ section
• Long videos (>1hr): Process in chunks; the script outputs ~500-800 lines for a typical 1.5hr video
• Non-English videos: The transcript will be in the video's language; summarize in the output language per the rules above (default 中文)