chubbyguan/chubbyskills

What it solves

Chubby Skills addresses the fragmentation of content consumption by providing a unified pipeline to capture, organize, and retrieve information from various Chinese and global social platforms. It transforms ephemeral social media feeds into a structured, local knowledge base that can be directly queried and utilized by AI Agents, preventing valuable information from being lost in the "feed" and turning it into reusable knowledge assets.

How it works

The project operates across three distinct layers:

1. Capture Layer: Uses specialized "skills" to scrape and transcribe content from platforms like Bilibili, YouTube, TikTok, Douyin, X (Twitter), WeChat Official Accounts, and podcasts, converting them into a standardized Markdown format with detailed metadata (Schema v1).
2. Knowledge Layer: Integrates these Markdown files into a local vault (e.g., Obsidian), providing indexing, full-text search, and semantic retrieval capabilities using either lightweight local embeddings or providers like OpenAI.
3. Agent Layer: Implements an MCP (Model Context Protocol) server, allowing AI Agents (such as Claude Code or Codex) to search, read, and interact with the local knowledge base to complete tasks based on the user's captured data.

Who it’s for

• Content Creators: To build a repository of viral notes, videos, and articles for topic research.
• Learners: To convert video and audio content into text, flashcards, and structured notes.
• Researchers: To maintain a local, agent-accessible reference library.
• Agent Users: To build automated workflows combining content ingestion and AI retrieval.

Highlights

• Multi-Platform Support: Comprehensive coverage of 10+ platforms including Bilibili, YouTube, X, and WeChat.
• Local-First Privacy: Entirely local execution and storage, ensuring data privacy.
• Transcription Capabilities: Built-in support for video and podcast transcription using tools like FunASR and Faster-Whisper.
• Agent Integration: Full MCP server implementation for seamless AI Agent access to personal knowledge.
• Content Enrichment: Optional AI-powered summarization, tagging, and value judgment via DeepSeek API.
• Standardized Output: Strict Markdown frontmatter and schema for consistent data portability.