YouTube Summary com Extração de Slides
ParthGanatra/agent-skills
Resumos estruturados de YouTube com slides extraídos e transcrição de conteúdo visual
Ver no GitHubSkill para Claude Code que transforma vídeos do YouTube em notas estruturadas (TL;DR, takeaways, capítulos) com extração opcional de slides. Detecta mudanças de slide via visão computacional, transcreve conteúdo (diagramas, código, gráficos) e embute as imagens principais na nota. Integra com Obsidian para salvar automaticamente ou imprime na conversa.
Instalação
npx convez-ai ParthGanatra/agent-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Resumir palestras e talks técnicas em notas estruturadas
- Extrair e organizar slides de vídeos educacionais em Obsidian
- Recuperar conteúdo quando a transcrição automática do YouTube é incompleta
Veredito

Ferramenta bem construída e prática para quem assiste muito conteúdo em vídeo e quer transformá-lo em documentação. A extração de slides é diferencial real — salva trabalho manual em talks visuais. Exige `yt-dlp` e `ffmpeg` localmente; funciona melhor em macOS mas é portável.
README
A tradução para o português deste README ainda está na fila.
agent-skills
Agent skills for Claude Code, by @ParthGanatra.
Skills
youtube-summary
Turn a YouTube video into a structured note — not a raw transcript dump, but a real summary (TL;DR, key takeaways, read-more links, open questions), with the video's own chapters used as the section structure when it has them. Works out of the box (prints the note right in the conversation); point it at an Obsidian vault to save it there instead.
Its distinctive feature: opt-in slide extraction. For slide-heavy talks, it downloads the video, detects unique slides via crop-aware change detection, transcribes their content (diagrams, code, charts) by vision, and embeds the key slides inline in the note next to the prose. This also rescues summaries when the auto-caption transcript is truncated — the slides carry the rest.
- Transcript + metadata + chapters via
yt-dlp, with a cookie cache so YouTube auth/bot-checks only prompt once. - Slides via
ffmpeg(scene/mpdecimatechange detection on a clean slide crop). - Vault location is configurable — nothing is hardcoded.
Requires: yt-dlp, and (for slides) ffmpeg + ffprobe. macOS-oriented
(uses the open obsidian:// URI and --cookies-from-browser).
Install
Via the skills.sh CLI:
npx skills add ParthGanatra/agent-skills/youtube-summary
Or manually — copy the skill folder into your Claude Code skills directory:
git clone https://github.com/ParthGanatra/agent-skills
cp -r agent-skills/youtube-summary ~/.claude/skills/
Configure (optional, for youtube-summary)
Works out of the box: with no config, the note is printed in the conversation (no file written). If you ask for slides, it saves the note + images to the current directory instead, since the images need to live on disk.
To save into your Obsidian vault instead (and embed slides as ![[wikilinks]]),
tell the skill where the vault is:
mkdir -p ~/.config/youtube-summary
cp ~/.claude/skills/youtube-summary/config.example.sh ~/.config/youtube-summary/config.sh
$EDITOR ~/.config/youtube-summary/config.sh # set vault name, path, notes subfolder
(Env vars OBSIDIAN_VAULT_NAME / OBSIDIAN_VAULT_PATH / NOTES_SUBFOLDER
override the file for one-off runs.)
Use
In Claude Code:
/youtube-summary <url>— or just paste a URL and ask for a summary.- Add "this one has slides" / "include the slides" to trigger slide extraction for that specific video (it's off by default — slides mean a full video download).