imraywang/wewrite

What it solves

WeWrite streamlines the entire content creation process for WeChat Official Accounts. It eliminates the friction of manually brainstorming topics, gathering evidence, writing drafts, editing for quality, and formatting for the specific constraints of the WeChat ecosystem.

How it works

The project uses a three-layer decoupled architecture:

1. Prompt Layer: A set of 10 modular "skills" that handle the intellectual work (topic selection, fact-checking, persona-based writing, and reviewing).
2. Runtime Layer: A Python CLI (wewrite) that handles deterministic tasks like converting Markdown to WeChat-compatible HTML, calling APIs for image generation, and pushing drafts to the WeChat backend.
3. State Layer: A local directory (~/.wewrite/) that stores user configurations, writing styles, historical data, and learned preferences.

It follows a strict pipeline: Hotspot analysis $\rightarrow$ Topic scoring $\rightarrow$ Task brief $\rightarrow$ Evidence gathering $\rightarrow$ Draft $\rightarrow$ Editorial review $\rightarrow$ Final copy. Users can then optionally trigger independent modules for AI image generation, layout preview, or publishing.

Who it’s for

• WeChat Official Account creators who produce frequent hot-topic, educational, or storytelling content.
• Users of AI Agents (like Claude Code, Codex, OpenClaw, or Hermes) who want a specialized content creation toolkit.
• Content creators looking to repurpose a single article for other platforms like Xiaohongshu or Douyin.
• Writers who want an AI to gradually learn and mimic their personal writing style over time.

Highlights

• End-to-End Workflow: Handles everything from trend spotting (via Weibo/Baidu) to draft pushing.
• Quality Control: Implements a "Task Brief" and "Claims vs. Evidence" system to prevent AI hallucinations and ensure factual accuracy.
• Persona-Driven: Includes 7 pre-defined writing personas (e.g., "Cold Analyst", "Midnight Friend") to control tone and rhythm.
• Advanced Layout Engine: 18+ built-in WeChat-compatible themes with support for dark mode and a learn-theme feature to extract styles from existing URLs.
• Style Flywheel: A learning mechanism that updates a personal playbook based on user edits and exemplar articles.
• Multi-platform Adaptation: True content-level rewriting for different social media formats rather than simple summaries.