I’ve been using these AI tools for half a year, so I decided to open-source them. They’re not big projects — just scripts I use daily. Transcribing Douyin and Bilibili videos, turning podcasts into text, processing WeChat articles, scanning industry intel… I’ve written 11 in total. I used them privately for a while, recently organized them, added installation scripts and documentation, and put them on GitHub. If you’re using agents like Claude Code, Codex, Hermes, or OpenClaw, you can install them directly. Or you can run them via command line.
Address: http://github.com/chubbyguan/chubbyskills…
Grab them if you need. If you find them useful, please give a star. If you find bugs, tell me.

chubbyguan/chubbyskills

Source: https://github.com/chubbyguan/chubbyskills

中文 · English

🧰 Chubby Skills

A set of AI Skills that collect Chinese omnichannel content into your personal knowledge base — I use them every day

License | Skills | Stars (https://github.com/chubbyguan/chubbyskills/stargazers) | Hype Weekly Chart (https://github.com/chubbyguan/chubbyskills) | Claude Code · Codex · OpenCode · OpenClaw · Hermes

I’m Chubby, an AI + e-commerce explorer.
I create content, build personal knowledge bases, and practice writing AI Agents/Skills. I’m used to automatically collecting the good stuff I browse every day — videos, podcasts, WeChat articles, Xiaohongshu, X (Twitter) — into my knowledge base, so that information truly settles instead of being forgotten after reading. The tools in this repository are the ones I personally use daily in this workflow.

If you share my interests, feel free to chat:

• 🐦 X / Twitter (https://x.com/Chubbyguan)
• 💬 Jike (https://web.okjike.com/u/a876838d-d9a8-494b-9494-bb3410b77dd5)
• 📕 Xiaohongshu (https://www.xiaohongshu.com/user/profile/57c061626a6a696f5a70f9a8)
• 📰 WeChat Official Account: 关关不过

All skills have been running in my own projects for a while and have proven to save time, so I’m open-sourcing them. No fancy tricks — just a few practical tools.

Each Skill here is a structured instruction set that agents can load directly, following the open standard Agent Skills. Claude Code, Codex, OpenCode, OpenClaw, and Hermes can all install them.

✨ What it can do for you

• 📥 Omnichannel collection — Douyin / Bilibili / Xiaohongshu / WeChat Official Account / X / Podcasts / YouTube, just drop the link
• 🎬 Auto-classification of image-text / video — image-text notes save images, video notes transcribe into text — no manual work
• ⚡ Subtitle priority, GPU-free — Instant text extraction when subtitles are available (YouTube / Bilibili), no need to install heavy models first
• 🧠 Settle into knowledge base — Unified format into Obsidian, plus an MCP server so any agent can query your library
• 🧩 Each skill independently installable — Install only the one you need; pure image-text / text collection has zero dependencies

In a nutshell: Turn Chinese omnichannel content into your own searchable second brain.

📋 Table of Contents

Video Transcription

Podcast Transcription

Content Processing

Content Enrichment

Knowledge Base Management

Workflows

📦 Installation

Method 1: One-command install (recommended)

git clone https://github.com/chubbyguan/chubbyskills.git
cd chubbyskills
bash setup.sh            # Install all dependencies
bash setup.sh podcast    # Install only a specific skill

Method 2: Manual install

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

Method 3: Agent install

In agents that support Skills like Claude Code, Codex, OpenClaw, Hermes, just say:

Help me install this skill: https://github.com/chubbyguan/chubbyskills/tree/main/

Gitee mirror: https://gitee.com/chubbyguan/chubbyskills

✨ Skills

🎬 douyin-transcribe

“I used to spend ages fiddling with cookies and yt-dlp to transcribe Douyin videos. Now it’s done in one sentence.”
Douyin video → download audio → SenseVoice-Small transcription → save as Markdown.

Credits: vangie/douyin-transcriber · FunAudioLLM/SenseVoice
→ SKILL.md · Scripts

📺 bilibili-transcribe

“Finally I can turn all those informative Bilibili videos into text and read them at my own pace.”
Bilibili video → yt-dlp download audio → SenseVoice-Small transcription → save as Markdown.

Credits: yt-dlp/yt-dlp · FunAudioLLM/SenseVoice
→ SKILL.md · Scripts

🎵 tiktok-transcribe

“Great content on TikTok can now be saved and read later.”
TikTok video → download audio → SenseVoice-Small transcription → save as Markdown. Supports vm.tiktok.com short links.

Credits: yt-dlp/yt-dlp · FunAudioLLM/SenseVoice
→ SKILL.md · Scripts

📱 weibo-transcribe

“Finally, video content on Weibo can be turned into text.”
Weibo video → download audio → SenseVoice-Small transcription → save as Markdown. Supports weibo.com and m.weibo.cn.

Credits: yt-dlp/yt-dlp · FunAudioLLM/SenseVoice
→ SKILL.md · Scripts

💡 zhihu-transcribe

“Video answers on Zhihu can now be transcribed and saved.”
Zhihu video → download audio → SenseVoice-Small transcription → save as Markdown.

Credits: yt-dlp/yt-dlp · FunAudioLLM/SenseVoice
→ SKILL.md · Scripts

🌍 youtube-transcribe

“I can finally understand English YouTube content easily.”
YouTube video → yt-dlp download → SenseVoice-Small transcription → auto-translate English to Chinese → output bilingual Markdown.

Credits: yt-dlp/yt-dlp · FunAudioLLM/SenseVoice · DeepSeek
→ SKILL.md · Scripts

🎙️ podcast-transcribe

“Can’t finish listening to podcasts? Turn them into text and read whenever you want.”
Podcast audio → download → faster-whisper transcription → save as Markdown. Supports Xiaoyuzhou and Ximalaya; supports RSS batch download.

Credits: SYSTRAN/faster-whisper · OpenAI Whisper
→ SKILL.md · Scripts

📰 wechat-article-ingest

“WeChat Official Account articles become structured knowledge directly.”
WeChat Official Account article → Markdown extraction → Layer A opinion extraction + Layer B question chain generation. Supports direct link fetching.

Credits: microsoft/markitdown · pymupdf/PyMuPDF · dontbesilent
→ SKILL.md · Scripts

📕 xiaohongshu-ingest

“Collect viral notes and get ready-to-write topic suggestions on the fly.”
Xiaohongshu note → unified frontmatter Markdown (title/body/tags/author/likes/saves/comments) → DeepSeek viral dissection (audience × scenario × pain point × emotion × hook) → 5 derivative topics. Auto-classifies image-text/video: image-text downloads images locally; video extracts direct links and transcribes to text (same as Douyin). Supports xhslink short links; recommended to set XHS_COOKIE to avoid anti-crawling.

Credits: DeepSeek · Xiaohongshu viral methodology
→ SKILL.md · Scripts

🐦 x-ingest

“One tweet: save images, transcribe videos.”
X tweet → unified frontmatter Markdown (body/author/likes/replies/topics). Auto-classifies image-text/video: image-text downloads images locally; video extracts highest-bitrate mp4 direct link and transcribes to text. Uses X’s official embed endpoint — no login, no API key. Currently supports single tweets.

Credits: FunAudioLLM/SenseVoice · X official embed endpoint
→ SKILL.md · Scripts

✨ content-enrich

“One-click enrichment of raw collected text into usable knowledge with summaries, key points, and tags.”
Automatically appends meta information to any collected output (transcripts / articles / notes): uses DeepSeek to extract a one-line summary, 3-5 key points, domain, tags, and a “worth deep reading?” judgment. Writes into frontmatter and inserts a ## 📝 Summary block at the top of the body; original content fully preserved. Supports single-file in-place enhancement and whole-directory batch processing; idempotent and re-runnable. This is the processing layer that connects “collection” with “knowledge base.”

Credits: DeepSeek
→ SKILL.md · Scripts

🧠 knowledge-base-management

“From material ingestion to health checks — a complete workflow for your knowledge base.”
Full lifecycle management of an Obsidian knowledge base: three-tier architecture (material library / Wiki / output), ABC grading of materials, health checks and cleanup, GBrain/GraphRAG/LLM Wiki integration, directory organization and archiving. Comes with an MCP Server (mcp_server.py) — exposes knowledge base search to any MCP agent, completing the “collect and write + MCP query” loop.

Core capabilities:

• 📥 Material ingestion: ABC grading + auto-sync from WeChat Official Account / group chats
• 🔍 Health checks: broken link repair, frontmatter completion, deduplication and archiving
• 🛠️ Tool integration: GBrain search + GraphRAG discovery + LLM Wiki writing
• 📂 Directory organization: full vault audit, batch archiving, file naming conventions

Credits: Obsidian · GBrain · GraphRAG
→ SKILL.md · Scripts

📡 industry-intelligence-radar

“Know early = Act early = Benefit early”
Multi-source intelligence scanning system: X/Twitter + Jike + V2EX + Hacker News + 36kr → keyword filtering → trend detection → daily intelligence brief.

Core capabilities:

• 🔍 Multi-source parallel scanning: X, Jike, V2EX, HN, 36kr
• 🏷️ Smart filtering: deduplication, timeliness, signal strength grading
• 📈 Trend detection: sudden/ongoing/emerging trend identification
• 📋 Intelligence brief: high-signal events + trend observations + opportunity insights

Covered domains: AI/Agent, semiconductors, aerospace, new energy, gaming, cross-border e-commerce, entrepreneurship/investment
→ SKILL.md · Scripts

📚 learning-notes-automation

“Watching a video ≠ learning; generating flashcards = remembering.”
Automated learning content processing: video/podcast transcription → knowledge point extraction → flashcard generation → knowledge graph update.

Core capabilities:

• 🎬 Multi-source input: YouTube, Bilibili, podcasts, Douyin, WeChat Official Account
• 📝 Knowledge point extraction: concepts, facts, methodologies, quotable quotes
• 🃏 Flashcard generation: concept cards, Q&A cards, comparison cards, step cards (Anki-compatible)
• 🕸️ Knowledge graph: entity extraction + relationship mapping + auto-ingestion

Output formats:

• Anki flashcard file (importable directly)
• Learning note Markdown
• Knowledge base entries

→ SKILL.md · Scripts

🔧 Environment Requirements

💡 Not sure what to install? Run python3 tools/check_env.py first — it will tell you which dependencies are missing and which features are zero-dependency (lightweight mode).

One-command install

bash setup.sh            # All dependencies
bash setup.sh podcast    # Only a specific skill

Or manually install the requirements.txt in each skill’s directory.

Video transcription (Douyin/Bilibili/TikTok/Weibo/Zhihu/YouTube)

pip install -r douyin-transcribe/requirements.txt

# System dependencies
brew install ffmpeg yt-dlp               # macOS
# sudo apt install ffmpeg && pip install yt-dlp   # Ubuntu

# YouTube translation (optional)
export DEEPSEEK_API_KEY=your-key

Podcast transcription

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

WeChat article processing

pip install -r wechat-article-ingest/requirements.txt

Xiaohongshu collection

# Collection and dissection are both zero-pip-dependency (stdlib only)
export XHS_COOKIE="your xiaohongshu cookie"   # Optional, improves collection success rate
export DEEPSEEK_API_KEY="your-key"            # Required for viral dissection

X/Twitter collection

# Image-text/text collection: zero dependencies, no login; video transcription requires funasr + ffmpeg
pip install funasr modelscope torch torchaudio   # Only needed for video tweets

Content enrichment

# Zero pip dependency (stdlib only)
export DEEPSEEK_API_KEY="your-key"   # For summary/key points/tag extraction

Knowledge base management

# The health check script vault_health_check.py is zero-dependency, pure stdlib — no install needed
# The following are optional third-party tools (not provided with this repo):
# GBrain (knowledge base search) — pip install gbrain
# GraphRAG (knowledge graph discovery) — separate project, see SKILL.md
# LLM Wiki — integrate into Obsidian vault

Industry intelligence radar

# scan.py is zero-dependency, pure stdlib (HN + V2EX + RSS) — no install, no API key required
# X/Jike signals are supplemented by the agent's web search (optional)

Learning notes automation

# make_notes.py requires DeepSeek API key for knowledge point extraction and flashcard generation
export DEEPSEEK_API_KEY=your-key

# Transcription part reuses this repo's *-transcribe skills (see "Video transcription" dependencies above)

📐 Unified frontmatter convention

All collection skills produce Markdown with a unified frontmatter, making it easy to aggregate and query by source/platform in a knowledge base:

Video/audio types may also carry language and translated fields. The platform field is machine-readable; it’s recommended to use it in Obsidian/Dataview for aggregated views by platform.

⚖️ Compliance & Disclaimer

All collection skills (video / podcast / WeChat Official Account / Xiaohongshu / X, etc.) in this repository are intended for personal learning and research purposes only:

• Please comply with each target platform’s Terms of Service (ToS), robots.txt, and applicable laws and regulations
• Control request frequency — do not use for mass scraping, commercial crawling, redistribution, or any activity that infringes on others’ rights
• Copyright of scraped/transcribed content belongs to the original authors; for citations and republications, obtain permission and provide attribution
• When login state (e.g., cookies) is involved, use it only within the scope of your own account and your own authorized usage
• This project is provided “AS IS”; the author assumes no responsibility for any consequences arising from the use of these tools

If a platform or rights holder believes a skill is inappropriate, feel free to raise an issue — I will address it promptly.

🙏 Acknowledgments

Repository structure inspiration

• KKKKhazix/khazix-skills — reference for repo structure and README style

Speech recognition

• FunAudioLLM/SenseVoice — Alibaba’s open-source Chinese speech recognition model
• SYSTRAN/faster-whisper — CTranslate2 implementation of Whisper
• OpenAI Whisper — groundbreaking speech recognition model

Video downloading

• yt-dlp/yt-dlp — powerful video downloader supporting thousands of platforms
• vangie/douyin-transcriber — inspiration for Douyin download approach

Document processing

• microsoft/markitdown — document to Markdown converter
• pymupdf/PyMuPDF — PDF processing library

Knowledge base management

• Obsidian — knowledge base platform
• GBrain — knowledge base search tool
• GraphRAG — knowledge graph discovery
• Karpathy LLM Wiki — LLM Wiki writing pattern

AI Agent & translation

• Agent Skills — open standard for Agent Skills
• DeepSeek — LLM for translation
• dontbesilent — A+B dual-track processing methodology
• Hermes Agent — runtime environment for these skills

Thank you to all open-source contributors! 🙏

🌟 About

These skills are all things I use daily. If open-sourcing them helps you, just give a ⭐.

• 🐦 X / Twitter (https://x.com/Chubbyguan)
• 💬 Jike (https://web.okjike.com/u/a876838d-d9a8-494b-9494-bb3410b77dd5)
• 📕 Xiaohongshu (https://www.xiaohongshu.com/user/profile/57c061626a6a696f5a70f9a8)
• 📰 WeChat Official Account: 关关不过

MIT License · Free to use / modify / redistribute

Made by @chubbyguan