/youtube

YouTube Transcript Extraction

I'll extract the transcript from a YouTube video.

Step 1: Parse and validate input

Parse $ARGUMENTS to extract the YouTube URL or video ID. Do NOT use bash conditionals — handle all logic in your reasoning layer.

1.1: Check for empty input

• If $ARGUMENTS is empty or whitespace only:
  • Show error: "No YouTube URL or video ID provided"
  • Show message: "Usage: /youtube {youtube-url-or-video-id}"
  • Exit gracefully

1.2: Determine URL and video ID

• If $ARGUMENTS is an 11-character alphanumeric string (video ID only):
  • Construct full URL: https://www.youtube.com/watch?v={VIDEO_ID}
  • Store as VIDEO_URL and VIDEO_ID
• If $ARGUMENTS contains a YouTube URL:
  • From youtube.com/watch?v=XXXXXXXXXXX — extract the v parameter as VIDEO_ID
  • From youtu.be/XXXXXXXXXXX — extract the path segment as VIDEO_ID
  • From youtube.com/shorts/XXXXXXXXXXX — extract the path segment as VIDEO_ID
  • From youtube.com/live/XXXXXXXXXXX — extract the path segment as VIDEO_ID
  • Strip any query parameters or timestamps after the video ID
  • Store the original URL as VIDEO_URL
• If $ARGUMENTS doesn't match any known format:
  • Show error: "Invalid YouTube URL or video ID"
  • Show message: "Accepted formats: YouTube URL, youtu.be short link, or 11-char video ID"
  • Exit gracefully

Note: Parse the URL in your reasoning layer. Do NOT use bash if/elif/else conditionals.

Step 2: Create temp directory

Run: mkdir -p /tmp/yt-transcript

Step 3: Fetch video title

Run: yt-dlp --print title "VIDEO_URL"

• If the command succeeds: store the output as VIDEO_TITLE
• If the command fails with "Private video" or "Video unavailable":
  • Show error: "Video is private or unavailable"
  • Exit gracefully
• If the command fails with "not a valid URL" or similar:
  • Show error: "Invalid YouTube URL: VIDEO_URL"
  • Exit gracefully
• If yt-dlp is not found:
  • Show error: "yt-dlp not found. Install with: brew install yt-dlp"
  • Exit gracefully

Step 4: Fetch subtitles

Run: yt-dlp --write-auto-sub --sub-lang en --skip-download --sub-format vtt -o "/tmp/yt-transcript/%(id)s" "VIDEO_URL"

• If the command succeeds: continue to Step 5
• If the output contains "no subtitles" or indicates no subs available:
  • Show message: "No English subtitles available for this video"
  • Show the video title that was fetched
  • Show suggestion: "This video may not have captions, or captions may be in another language."
  • Exit gracefully

Step 5: Locate the VTT file

Run: ls /tmp/yt-transcript/VIDEO_ID*.vtt

• Store the first matching file path as VTT_FILE
• If no .vtt file found:
  • Show error: "Subtitle file not found after download"
  • Exit gracefully

Step 6: Parse VTT to clean text

Run the following sed + awk pipeline as a single command:

sed '/^WEBVTT/d; /^Kind:/d; /^Language:/d; /^NOTE/d; /^$/d; /^[0-9]*$/d; /-->/d; s/<[^>]*>//g; s/&nbsp;/ /g; s/&amp;/\&/g; s/&lt;/</g; s/&gt;/>/g' "VTT_FILE" | awk '!seen[$0]++' > "/tmp/yt-transcript/VIDEO_ID.txt"

This pipeline:

• Removes WEBVTT header, Kind/Language metadata, NOTE blocks
• Removes empty lines, numeric cue IDs, timestamp lines (containing -->)
• Strips HTML tags (YouTube auto-subs contain <c> tags)
• Decodes common HTML entities
• Deduplicates lines (YouTube auto-subs repeat lines across cues)

Step 7: Get character count

Run: wc -c < /tmp/yt-transcript/VIDEO_ID.txt

Store the trimmed output as CHAR_COUNT.

Step 8: Read and present the transcript

Use the Read tool to read /tmp/yt-transcript/VIDEO_ID.txt.

Present the output in this format:

**VIDEO_TITLE**
VIDEO_URL
CHAR_COUNT characters | Saved to /tmp/yt-transcript/VIDEO_ID.txt

---

TRANSCRIPT_CONTENT

Step 9: Clean up VTT file

Run: rm -f "VTT_FILE"

Keep the .txt file for later reference during the session.

Error Handling:

• If yt-dlp is not found: "Install yt-dlp: brew install yt-dlp"
• If video is private/unavailable: Show clear error with the URL
• If no English subs: Suggest the video may not have captions
• If VTT parsing produces empty output: Show warning that subtitles may be in a non-standard format