chubbyguan/chubbyskills

Chubby Skills – A Local Knowledge‑Base Toolkit for Creators

What it is

• A collection of 14 Agent Skills and a command‑line interface that lets you capture videos, podcasts, articles and local documents, turn them into searchable Markdown files, and export evidence‑backed material packs that agents can consume.
• Works entirely offline for the core workflow – no third‑party packages, API keys or model downloads are required unless you enable optional transcription or enrichment features.

Key capabilities

Area	What you can do
Capture	Grab subtitles from B‑site/YouTube, transcribe TikTok/Douyin/Weibo/知乎 videos, pull X/Twitter posts, scrape WeChat public articles, import local Markdown/TXT/PDF (text‑layer only).
Import & Index	Convert everything to Markdown with front‑matter (title, source URL, author, timestamps, likes, etc.), store attachments locally, and build a searchable index (keyword + optional semantic‑lite).
Search & Retrieval	Find snippets by keyword (or semantic search if you install the optional vector store), see the original line numbers, source URL and a SHA‑256 hash for verification.
Export	Generate a brief – a Markdown/JSON package containing the exact excerpts, their line numbers, source metadata and a hash, ready for an LLM agent to use for drafting or topic selection.
Subscription	Set up RSS/YouTube‑channel or podcast subscriptions that periodically pull new items, queue them for ingestion, and optionally produce daily digests (no LLM needed, optional DeepSeek summarisation).
Agent integration	Install the knowledge‑base skill package into agents such as Claude Code, OpenCode, etc., or run a lightweight MCP server that offers search, retrieval and indexing APIs.

How to get started (quick‑run)

1. Clone the repo and initialise a vault:

   git clone https://github.com/chubbyguan/chubbyskills.git
   cd chubbyskills
   python3 tools/chubby.py init --vault "$PWD/creator-vault"
   

2. Capture a public X post (no extra deps needed):

   python3 tools/chubby.py ingest "https://x.com/OpenAI/status/1663696190960173056" --no-enrich
   python3 tools/chubby.py search "ChatGPT"
   

3. Export a source‑backed brief:

   python3 tools/chubby.py brief --topic "ChatGPT" --output "$PWD/creator-vault/30_Output/brief.md"
   

   The resulting brief.md contains the exact excerpt, its line numbers, the original URL and a SHA‑256 hash.

Installation options

• Zero‑dependency CLI – run python3 tools/chubby.py … directly after cloning (Python 3.11+).
• Editable package – pip install -e . gives you a chubby command.
• Full skill install – python3 tools/install_skill.py --all --dest /path/to/agent/skills copies the 14 skill directories into an agent’s skill folder.
• Optional extras – bash setup.sh light (checks Python only), bash setup.sh video (adds ffmpeg and local transcription models), bash setup.sh podcast (adds funasr), or install yt-dlp, pymupdf, etc., as listed in the Platform capabilities table.

Typical use cases

• Content creators who need to keep a personal archive of source material (tweets, YouTube videos, podcasts) with full citation data for later writing.
• Researchers / knowledge workers building a local evidence base that can be queried by LLM agents without sending raw content to the cloud.
• Agents that require a searchable knowledge base – the MCP server exposes search, read, index and other utilities for seamless integration.

Limitations & caveats

• Platform scraping respects the site’s public APIs; it does not bypass anti‑scraping measures. Some sites may return only previews (e.g., X long articles) unless you provide valid cookies.
• Video/audio transcription relies on external tools (ffmpeg, funasr, optional Qwen‑ASR) and may need to download a model on first run.
• Semantic search and content enrichment (summaries, tags) are optional and require API keys (DeepSeek, OpenAI, etc.) – they are not part of the core offline flow.
• PDF import extracts only the text layer; scanned PDFs need OCR beforehand.
• The tool is intended for personal, non‑commercial use. Users must respect copyright and platform terms when storing and re‑using content.

Where to learn more

• Quick‑start guide and detailed docs are under the docs/ folder (installation, creator workflow, subscription, MCP configuration, platform fallbacks, etc.).
• Example outputs are in examples/README.md.
• Community discussions, bug reports and feature requests live in the repository’s Discussions and Issues sections.

---

Chubby Skills bridges the gap between raw internet content and a structured, citation‑rich knowledge base that can be fed directly to AI agents, all while keeping the data on your own machine.