🧬 yt-ribosome
Run Transcription and Translation for Youtube or playlist into clean Markdown transcripts.
transcription → translation: pull the spoken words out of a video, then localize them.
Video is great, but for learning/information purposes, it's too less dense.
In the world where there's just so much information out there, you might want to optimize
your information ingestion time, and automated transcription --> translation pipeline help you
get any youtube knowledge out there efficiently.
New in v0.2.0: the full-blog skill produces a self-contained HTML page with the
transcript and ~10–25 meaningful frame snapshots embedded inline. Click any frame to
jump to that moment on YouTube. Translate the result with translate to ship localized
blog posts.
YouTube URL ──▶ transcribe ──▶ Markdown (original language) ──▶ translate ──▶ Markdown (your language)
Why
Getting usable text out of YouTube is fiddly: captions are sometimes missing, come in the wrong
language, are riddled with rolling-duplicate lines, or are auto-translated rather than original.
And once you have a transcript, translating a long talk hits model output limits and mangles
formatting. yt-ribosome handles all of that for you, as natural-language skills inside Claude Code.
Features
- 🎯 Original-language transcripts — pulls the true source ASR track, not an auto-translation.
- 🔁 Quality-first transcription — uses creator-uploaded manual subtitles when present;
otherwise transcribes the audio with the OpenAI API (
gpt-4o-transcribe). YouTube auto-captions are skipped by default; opt back in with --allow-auto-captions.
- 🧹 Clean output — strips YouTube's rolling-duplicate caption lines; paragraphs even
punctuation-less captions; writes
.srt too when timestamps exist.
- 🌍 Translate to any language — OpenAI or Gemini, preserving Markdown structure and URLs.
- ✍️ Transcription-error aware — the translator silently fixes ASR slips (e.g. "Cloud" → Claude).
- ⏭️ Skips what's already done — files already in the target language are passed through; both
steps are resumable.
- ⚡ Parallel — videos and files are processed concurrently.
Full-Blog Output
full-blog produces a self-contained HTML page with the transcript and meaningful
frames embedded inline at the timestamp each frame was captured. Each <figure> deep-
links back to YouTube via ?t= so readers can jump to the moment.
# v0.2.0 — turn a YouTube URL into an HTML blog with images
python3 skills/full-blog/scripts/full_blog.py "<URL>" --out-dir blogs
# Then translate (HTML-aware):
python3 skills/translate/scripts/translate.py blogs/ --to Korean
Output structure:
blogs/
├── 01 - My Talk.html
├── 01 - My Talk/
│ ├── 00_03_12.jpg
│ └── 00_05_44.jpg
└── _run_summary.json
Cost target: ~$0.10 per 60-min video with gemini-2.5-flash. ~$0.03 with gemini-2.0-flash.
Skills
| Skill | What it does | Say something like |
|---|
transcribe | YouTube → original-language Markdown | "transcribe this video/playlist", "유튜브 트랜스크립트 받아줘" |
translate | transcript/Markdown/HTML files → target language | "translate these to Korean", "한국어로 번역해줘" |
transcribe-and-translate | both, end to end | "transcribe this and translate it to Korean" |
full-blog | YouTube → HTML blog with embedded frame snapshots | "make a full blog from this video", "유튜브 영상을 블로그로 변환해줘" |
Install
/plugin marketplace add SSFSKIM/yt-ribosome
/plugin install yt-ribosome@yt-ribosome
Or run it locally without installing:
claude --plugin-dir /path/to/yt-ribosome
Prerequisites
| Requirement | Needed for |
|---|
yt-dlp (recent) + ffmpeg | all transcription |
openai + OPENAI_API_KEY | audio fallback and OpenAI translation |
google-genai + GEMINI_API_KEY or GOOGLE_API_KEY | Gemini translation |
imagehash, Pillow (pip install imagehash Pillow) | full-blog: perceptual-hash frame deduplication |
beautifulsoup4 (pip install beautifulsoup4) | translate: HTML support (v0.2.0) |
pip install -U yt-dlp openai google-genai # ffmpeg: brew install ffmpeg / apt install ffmpeg
API keys are read from the environment or a .env file in the current working directory. A Gemini
key starting with AQ. is auto-detected as a Vertex AI Express key; an AIza… key uses the
standard Gemini API.
Quickstart
In Claude Code, just ask:
"Transcribe https://youtube.com/watch?v=… and translate it to Korean."
