transcribe

name	transcribe
description	Transcribe specific video URL(s). Use when you have one or more TikTok video URLs to process.
user_invocable	true

Transcribe Specific Videos

This command processes specific TikTok videos by URL - no need to scrape an entire profile.

Workflow

Step 1: Get Video URLs

Ask the user:

Paste the TikTok video URL(s) you want to transcribe.

You can paste one URL or multiple URLs (one per line):

https://www.tiktok.com/@username/video/123456789
https://www.tiktok.com/@username/video/987654321

Parse the URLs from the user's message. Extract any URLs that match TikTok video patterns:

• https://www.tiktok.com/@username/video/ID
• https://tiktok.com/@username/video/ID
• https://vm.tiktok.com/SHORTCODE

Step 2: Verify Environment

Check that the project is set up:

cd /path/to/project
source .venv/bin/activate 2>/dev/null || echo "venv not found"

If venv doesn't exist, tell user to run /start first.

Step 3: Process Each Video

For each URL provided, run:

source .venv/bin/activate && python -c "
from short_form_scraper.scraper.tiktok import TikTokScraper
from short_form_scraper.downloader.video import VideoDownloader
from short_form_scraper.transcriber.whisper import WhisperTranscriber
from pathlib import Path

# Initialize
scraper = TikTokScraper('https://www.tiktok.com/@placeholder')  # Just for metadata fetching
downloader = VideoDownloader()
transcriber = WhisperTranscriber()

# Create directories
Path('transcripts').mkdir(exist_ok=True)
Path('state').mkdir(exist_ok=True)

# Video URL
video_url = 'VIDEO_URL_HERE'

print(f'Processing: {video_url}')

# Get metadata
metadata = scraper.get_single_video_metadata(video_url)
print(f'Title: {metadata.title}')
print(f'Duration: {metadata.duration}s')

# Download
audio_path = downloader.download(metadata.url, Path(f'state/audio_{metadata.id}'))
print(f'Downloaded audio: {audio_path}')

# Transcribe
transcript = transcriber.transcribe(audio_path)
print(f'Transcribed: {len(transcript)} characters')

# Save
transcript_file = Path('transcripts') / f'{metadata.id}.txt'
content = f'''Video ID: {metadata.id}
Title: {metadata.title}
URL: {metadata.url}
Duration: {metadata.duration}s

--- TRANSCRIPT ---

{transcript}
'''
transcript_file.write_text(content)
print(f'Saved: {transcript_file}')

# Cleanup audio
audio_path.unlink()
print('Done!')
"

Run this for each URL the user provided.

Step 4: Summarize Transcripts (Claude Code)

After transcripts are generated, YOU (Claude Code) will:

1. Read each new transcript file from transcripts/

2. Analyze and extract:

   • Topic: 2-4 word kebab-case topic name
   • Summary: One-sentence summary
   • Key Tips: 3-5 actionable bullet points
   • Details: Additional context

3. Create summary files in summaries/{topic}/{slugified-title}.md:

   IMPORTANT: Filename from Video Title

   • Extract the video title from the yt-dlp metadata (stored in transcript header as "Title: ...")
   • Convert to kebab-case slug: lowercase, spaces to dashes, remove special chars
   • Example: "How to Use Context Windows" → how-to-use-context-windows.md
   • Example: "Claude Code Tips & Tricks!" → claude-code-tips-tricks.md

   ---
   video_id: {id}
   title: {title}
   url: {url}
   topic: {topic}
   ---
   
   # {Title}
   
   ## Summary
   {one-sentence summary}
   
   ## Key Tips
   - {tip 1}
   - {tip 2}
   - {tip 3}
   
   ## Details
   {additional context}
   
   ## Full Transcript
   {original transcript}
   

4. Update INDEX.md with the new summaries

Step 5: Show Results

Display:

• Video title and URL
• Transcript location
• Summary location
• Key tips extracted

Example

User: /transcribe

Claude: What TikTok video URLs would you like to transcribe?

User:

https://www.tiktok.com/@example_creator/video/7597629199486029070
https://www.tiktok.com/@example_creator/video/7597614123362323767

Claude: [Processes each video, creates transcripts and summaries]

Handling Multiple URLs

When multiple URLs are provided:

1. Process them sequentially
2. Report progress: "[1/3] Processing..."
3. Continue even if one fails
4. At the end, report success/failure for each