中文 · English

🧰 Chubby Skills

Your feed forgets. Your knowledge base doesn't. — pull content from everywhere into your second brain

License Skills Stars Hype Weekly

Claude Code Codex OpenCode OpenClaw Hermes

These are AI Skills I've been using in my own projects. They've proven useful, so I'm open-sourcing them.

Each Skill here is a structured instruction set that Agents can load directly, following the Agent Skills open standard. Works with Claude Code, Codex, OpenCode, OpenClaw, and Hermes.

✨ What it does for you

  • 📥 Multi-platform ingest — Douyin, Bilibili, Xiaohongshu, WeChat, X, podcasts, YouTube; just drop a link
  • 🎬 Auto image/video routing — image notes keep their images, video notes get transcribed
  • ⚡ Subtitle-first, no GPU — instant text when YouTube/Bilibili captions exist
  • 🧠 Into your knowledge base — unified Markdown for Obsidian, plus an MCP server so any agent can query your vault
  • 🧩 Each skill installs independently — grab only what you need; image/text ingest is zero-dependency

In one line: turn Chinese content from everywhere into your own searchable second brain.


🚀 Quick Start

git clone https://github.com/chubbyguan/chubbyskills.git
cd chubbyskills
python3 tools/check_env.py

Install only what you need:

bash setup.sh          # light mode: text/image ingest + knowledge tools
bash setup.sh video    # ffmpeg + yt-dlp + funasr/torch stack
bash setup.sh podcast  # faster-whisper stack
bash setup.sh wechat   # WeChat/PDF extraction stack
bash setup.sh all      # everything

Run the one-command workflow:

python3 tools/chubby_ingest.py "https://www.bilibili.com/video/BVxxxx" -o output/
python3 tools/chubby_ingest.py "https://x.com/user/status/123" -o output/ --enrich
python3 tools/chubby_ingest.py "https://mp.weixin.qq.com/s/xxx" --vault ~/Documents/Obsidian/Inbox

Validate generated Markdown:

python3 tools/validate_outputs.py examples/outputs
python3 tools/validate_outputs.py output/

🧭 Capability Matrix

Capability Scope Dependencies Notes
Subtitle-first transcription Bilibili / YouTube yt-dlp Falls back to audio transcription when no captions exist
Video transcription Douyin / Bilibili / TikTok / Weibo / Zhihu / YouTube / XHS video / X video ffmpeg + funasr + torch Heavy first install; platform anti-bot rules may affect downloads
Podcast transcription Xiaoyuzhou / Ximalaya / RSS / local audio ffmpeg + faster-whisper Long episodes can take time
Image/text ingest Xiaohongshu / X / WeChat light or zero pip deps XHS works better with XHS_COOKIE
Enrichment Any Markdown output DEEPSEEK_API_KEY Adds summary, key points, tags
Knowledge base Obsidian vault health check is zero-dep; MCP needs mcp Set VAULT_DIR for MCP
One-command workflow common links and local audio/PDF uses matching skill deps Override with --skill when detection is ambiguous

📋 Table of Contents

Video transcription

Skill Platform One-liner
🎬 douyin-transcribe Douyin Video → transcribe → Markdown
📺 bilibili-transcribe Bilibili Video → subtitle-first transcribe → Markdown
🎵 tiktok-transcribe TikTok Video → transcribe → Markdown
📱 weibo-transcribe Weibo Video → transcribe → Markdown
💡 zhihu-transcribe Zhihu Video → transcribe → Markdown
🌍 youtube-transcribe YouTube Subtitle-first → transcribe → CN translation

Podcast

Skill Platform One-liner
🎙️ podcast-transcribe Xiaoyuzhou / Ximalaya Podcast → transcribe → Markdown (RSS batch)

Content ingest

Skill Platform One-liner
📰 wechat-article-ingest WeChat Article → Markdown + A/B insight extraction
📕 xiaohongshu-ingest Xiaohongshu Image saved / video transcribed + hook analysis
🐦 x-ingest X / Twitter Tweet → image saved / video transcribed (no login)

Enrich & knowledge base

Skill One-liner
✨ content-enrich Auto-add summary + key points + tags to any ingested note
🧠 knowledge-base-management Vault lifecycle + health check + MCP server
📡 industry-intelligence-radar Multi-source scan → daily intel brief
📚 learning-notes-automation Transcript → key points → Anki flashcards

📦 Installation

git clone https://github.com/chubbyguan/chubbyskills.git
cd chubbyskills
bash setup.sh          # light mode
bash setup.sh video    # video transcription stack
bash setup.sh podcast  # podcast transcription stack
bash setup.sh wechat   # WeChat/PDF extraction stack
bash setup.sh all      # everything

Option 2: Manual install

git clone https://github.com/chubbyguan/chubbyskills.git
cd chubbyskills
pip install -r requirements.txt            # All skills
pip install -r podcast-transcribe/requirements.txt   # Single skill

Option 3: Agent install

In any Agent that supports Skills (Claude Code, Codex, OpenClaw, Hermes, etc.), just say:

Install this skill: https://github.com/chubbyguan/chubbyskills/tree/main/<skill-name>

✨ Skills

🎬 douyin-transcribe

"Transcribing Douyin videos used to be a hassle with cookies and yt-dlp. Now it's one command."

Douyin video → download audio → SenseVoice-Small transcribe → save as Markdown. Supports short links and full links. No cookies, no login, no yt-dlp needed.

Why SenseVoice over Whisper

faster-whisper SenseVoice-Small
Chinese accuracy Average Exceeds Whisper
Speed (CPU) tiny: 149s/1h, small: 10-15min RTF ~0.04, 1h audio ≈ 2-3 min
VAD Extra setup needed Built-in fsmn-vad

What it does

  • 🔗 Supports Douyin short links (v.douyin.com/xxx) and full links
  • ⚡ Fast transcription: 11-minute video in just 22 seconds
  • 📝 Auto-generates Markdown with frontmatter
  • 🎯 Chinese accuracy exceeds Whisper, simplified Chinese output
  • 🔒 No login, no cookies, no API Key needed

→ SKILL.md · Scripts

🎙️ podcast-transcribe

"Can't finish listening to podcasts? Turn them into text."

Podcast audio → download → faster-whisper transcribe → save as Markdown. Supports Xiaoyuzhou, Ximalaya, and other platforms. RSS batch download supported.

What it does

  • 🎧 Supports Xiaoyuzhou, Ximalaya, and other podcast platforms
  • 📡 RSS batch download for entire podcast seasons
  • 📝 Auto-generates Markdown with timestamps
  • ⏱️ Resume support, skips already transcribed episodes
  • 📁 Auto-naming by episode number

Performance

Model Speed (CPU) Chinese Accuracy
faster-whisper tiny ~149s/1h Average
faster-whisper small ~10min/h Good (~85-90%)
faster-whisper large-v3 ~30-60min/h Best

→ SKILL.md · Scripts

📺 bilibili-transcribe

"So many great videos on Bilibili, now I can read them as text."

Bilibili video → yt-dlp download audio → SenseVoice-Small transcribe → save as Markdown. Supports BV IDs and full links. No login needed.

What it does

  • 📺 Supports Bilibili BV IDs and full links
  • ⚡ Fast transcription: 7-minute video in just 15 seconds
  • 📝 Auto-generates Markdown with frontmatter
  • 🎯 High Chinese accuracy, simplified Chinese output
  • 🔒 No login, no cookies needed

Triggers

Transcribe this Bilibili video: https://www.bilibili.com/video/BV1rrQGBeEen/
Help me transcribe this bilibili

Performance

Video Duration Transcription Time
5 min ~10s
10 min ~20s
30 min ~60s

🌐 Cross-platform: Claude Code · Codex · OpenCode · OpenClaw · Hermes

→ SKILL.md · Scripts


🔧 Requirements

Quick install

bash setup.sh          # All dependencies
bash setup.sh podcast  # Specific skill only

douyin-transcribe / bilibili-transcribe / youtube-transcribe

pip install -r douyin-transcribe/requirements.txt
# System: brew install ffmpeg yt-dlp  # macOS

podcast-transcribe

pip install -r podcast-transcribe/requirements.txt
# System: brew install ffmpeg  # macOS

🌟 About

I'm Chubby, an indie creator tinkering with AI and content. I like to auto-collect everything I come across — videos, podcasts, articles, Xiaohongshu, tweets — into my own knowledge base so it actually sticks instead of being forgotten. These skills are what I use daily for exactly that. If they help, give it a ⭐. Issues and discussions welcome.


MIT License · Free to use / modify / redistribute

Made by @chubbyguan

S
Description
chubby 自用知识库实用skill集合 - AI驱动的自动化工作流工具,可以帮助你从各个渠道获取信息,搭建自己的知识库
Readme MIT 264 KiB
Languages
Python 97.4%
Shell 2.6%