leter/zh-tech-writing

What it solves

It eliminates "AI-style" writing in Chinese technical documentation. Many LLMs produce text that is overly verbose, filled with marketing jargon, and uses repetitive sentence structures that feel unnatural to engineers. This project provides a set of rules and a workflow to ensure technical docs (READMEs, design docs, API specs, and tutorials) are concise, factual, and professional.

How it works

It is implemented as an "Agent Skill" compatible with tools like Claude Code. It guides the AI through a specific writing process:

1. Audience Definition: Identifies the target reader and the desired outcome.
2. Rule-Based Drafting: Applies strict rules for sentence length (under 20 characters per clause), active voice, and direct verbs.
3. AI-Style Scrubbing: Uses a 14-point "AI-style checklist" to identify and remove clichés, marketing adjectives, and translation-style phrasing.
4. Automated Formatting: Integrates with the autocorrect tool to automatically fix spacing between Chinese and English characters and correct punctuation.
5. Manual Refinement: Performs a final check on specific symbols like quotation marks and ellipses.

Who it’s for

Developers and technical writers who want their AI-generated Chinese documentation to sound like it was written by a human engineer rather than a generic AI assistant.

Highlights

• Fact-Driven: Replaces vague adjectives (e.g., "powerful," "seamless") with concrete facts, numbers, and error messages.
• Standardized: Based on established guidelines like Ruan Yifeng's Chinese Technical Documentation Style Guide.
• Tool Integration: Works with the skills CLI and supports automated punctuation correction via autocorrect.
• Customizable Workflow: Can be used to either directly rewrite files or provide a list of suggested improvements in a "Original → Modified → Reason" format.