WeChat Content Skill (wechat-content-skill)
This is an AI Agent Skill specifically designed for WeChat public account creators. It is not a simple toolbox, but a creative assistant with "deep thinking" capabilities.
It helps you complete the entire process from web-wide collection -> material organization -> deep writing, and has a built-in "reversal-driven" deep writing system, refusing to generate mediocre AI fluff.
Core Capabilities
1. Deep Research & Organization
- Auto-collection: Based on keywords, automatically searches for high-quality information sources from Google/Web (automatically filtering low-quality content).
- Smart library building: Automatically organizes collected information into
Topic(topic) andSource(source material), and stores them incontent-library. - Auto-indexing: Automatically maintains the local knowledge base through
index.json, getting smarter with use.
2. Deep Writing System
This is not simple "expansion", but creation based on "Reversal Thinking":
- Find an enemy: Every article must have an opposing side to avoid "correct nonsense".
- Create a reversal: The core logic is "you think it's A, but it's actually B".
- Generate 3 drafts at once: For the same topic, automatically generates 3 complete article drafts with different entry points for you to choose from.
Quick Start
1. Installation
Clone this project to the Skill management directory of your AI editor/IDE:
git clone https://github.com/CheeMao/wechat-content-skill.git
2. Loading and Usage
This project is compatible with editors that support Agentic Workflow:
Using in Antigravity (IDE)
- Ensure the Skills directory is loaded in Settings.
- Simply type commands in the dialog box to activate.
Using in Claude Code / Cursor / Trae
- Claude Code: Run
claudein the project root directory and just chat. - Cursor/Trae: Drag the entire folder as Context, and use it with Composer (Ctrl+I/Cmd+I).
Command Instructions
The Skill defines 4 standard work modes, which you can trigger via @wechat-content (or directly in natural language):
🎯 Mode A: Temporary Search
"I want to know the latest about Claude 3.5"
- Command:
@wechat-content search [keyword] - Function: Quick web search, extract core viewpoints, not saved to local library.
📥 Mode B: Material Accumulation (Core)
"Help me collect materials about Sora, I want to write an article"
- Command:
@wechat-content collect [keyword] - Function: Web search -> deep analysis -> automatically create local material files (saved in
content-library).
🔍 Mode C: Retrieve Materials
"Check if my library has any materials about large models"
- Command:
@wechat-content find [keyword] - Function: Retrieve existing materials from the local
content-library.
✍️ Mode D: Deep Creation (Core)
"Based on the collected Sora materials, help me write an article"
- Command:
@wechat-content write [topic keyword] - Function:
- Automatically retrieve related materials (automatically supplements from the web if insufficient).
- Automatically conceive 3 different "reversal" angles.
- Directly output 3 complete article drafts (including titles and body text).
- Execute silently throughout, directly giving results.
Directory Structure
wechat-content-skill/
├── SKILL.md # Core brain: defines all Prompts and workflows
├── content-library/ # Your local knowledge base (auto-generated, added to .gitignore)
│ ├── topics/ # Organized topic cards
│ ├── sources/ # Raw material segments
│ └── index.json # Knowledge base index
├── templates/ # Internal structural definitions (users do not need to manually modify)
└── scripts/ # Auxiliary scripts
FAQ
Q: Why do the generated articles sometimes feel very sharp?
A: SKILL.md has built-in rules for "removing AI flavor" and "finding an enemy", forcing the AI to seek conflict and reversal when writing, simulating the thinking of experienced self-media creators.
Q: Do I need to manually change files in the templates directory? A: No. Those are structural definitions for the AI to see. You only need to give the AI commands like "collect" or "write", and it will automatically handle all file operations.
License
MIT License


