Video Transcript Extraction

Seed question: How do I get the transcript of this video?

Extract text transcripts from any video source. Platform-aware — detects what tools are available and guides to the best method.

Setup Check

Before extracting, determine available tools:

Guide user to install what's missing for their platform if needed.

Determine Source and Method

1. YouTube URL?
   → Method A: Caption extraction (youtube-transcript-api)
   → Fallback: Whisper if no captions available

2. Other video URL?
   → Check if platform provides captions
   → Fallback: Whisper transcription

3. Local video/audio file?
   → Method C: Whisper directly
   → Apple Silicon: MLX-Whisper (faster)
   → Other: Whisper-cpp

Method A: YouTube Caption Extraction (Primary for YouTube)

CLI

youtube_transcript_api VIDEO_ID
youtube_transcript_api VIDEO_ID --languages en es
youtube_transcript_api VIDEO_ID --json > transcript.json

Python

from youtube_transcript_api import YouTubeTranscriptApi

transcript = YouTubeTranscriptApi.get_transcript("VIDEO_ID")
text_only = " ".join([t['text'] for t in transcript])

# List available languages
transcript_list = YouTubeTranscriptApi.list_transcripts("VIDEO_ID")

Bulk Processing

for vid in VIDEO_ID_1 VIDEO_ID_2 VIDEO_ID_3; do
  youtube_transcript_api "$vid" --json > "transcripts/${vid}.json"
  sleep 2
done

Method B: Whisper Transcription (Any Video)

MLX-Whisper (Apple Silicon)

mcp__mlx-whisper__transcribe_youtube(
  url="VIDEO_URL",
  language="en",
  task="transcribe",
  keep_file=true
)

Whisper-cpp (Cross-platform)

mcp__whisper-cpp__transcribe(
  file_path="/path/to/video.mp4",
  language="en"
)

Method C: Web Services (No Setup Required)

Extraction Workflow

1. CHECK: Does video have existing captions?
   → Try youtube-transcript-api first (fastest, zero compute)
   → Check language availability

2. IF NO CAPTIONS: Use Whisper
   → Apple Silicon: MLX-Whisper MCP
   → Other: Whisper-cpp MCP

3. IF TOOLS UNAVAILABLE: Web services
   → youtubetotranscript.com
   → Manual from YouTube UI

4. PROCESS: Clean transcript
   → Remove timestamps if not needed (saves tokens)
   → Save to file for downstream analysis

Token Optimization

Transcripts can be long. Strategies:

• Remove timestamps: " ".join([t['text'] for t in transcript]) — saves ~30% tokens
• Chunk processing: Split into 10K character chunks for analysis
• Summarize first: Quick summary pass before deep analysis
• Consolidated output: For 5+ videos, create all_transcripts.md with --- separators

Troubleshooting

Budget Mode

Activation (any of these):

1. Explicit flag: --budget or -b
2. Auto-detect: If budget-mode skill is active in session
3. Inherited: If calling skill passed budget context

When active:

• Prefer caption extraction over Whisper (lower compute cost)
• Skip timestamp inclusion by default
• Output text-only (no structured JSON)

Note: After context compaction, auto-detection may fail. Re-invoke budget-mode skill or pass --budget explicitly.

Cross-References

• Called by: youtube-research (Phase 2)
• Standalone: Invoke directly for any video transcription need
• Downstream: Extracted transcripts feed into youtube-research analysis or DIP investigations requiring video source material

Vasana

A vasana is a pattern that persists across unrelated contexts. If during this task you notice such a pattern emerging, it may be worth capturing. This skill works best alongside the vasana skill and vasana hook from the Vasana System plugin.

Modify freely. Keep this section intact.