KKKKhazix/human-writing

What it solves

It addresses the "AI-style" writing common in Chinese LLM outputs, where text is fluent but lacks a distinct human personality, often relying on repetitive structures, "report-style" phrasing, and predictable rhetorical patterns (such as the "not A, but B" trope).

How it works

The project provides a comprehensive framework for AI agents to produce more natural, human-like Chinese prose. It operates through a three-step process:

1. Material Validation: Ensures there is enough factual or narrative substance before writing to avoid filler text.
2. Writing Guidelines: Focuses on factual accuracy for non-fiction and causal logic for fiction, ensuring every paragraph introduces new information and utilizing plain language with natural rhythms.
3. Automated Review: A Python-based check script (check_prose.py) and a set of hard rules analyze the draft to remove repetitive explanations, excessive colons, and common AI buzzwords or rhetorical patterns.

Who it’s for

Content creators writing for platforms like Zhihu, WeChat Official Accounts, blogs, forums, and those producing scripts for oral broadcasts, novels, or popular science articles in Chinese.

Highlights

• Pattern-based Detection: Moves beyond simple keyword banning to detect structural "postures" (e.g., preventing the AI from creating a fake misunderstanding just to debunk it).
• Agent Integration: Designed as a "Skill" that can be installed directly into AI agents.
• Comprehensive Reference Library: Includes specific guidelines for different formats, including forum prose, fiction, and oral delivery.
• Lite Version: Provides a distilled version for users who want to paste instructions directly into chat windows like ChatGPT or Tongyi Qianwen.