This project addresses the pervasive issue of unnatural, boilerplate-heavy Chinese text generated by AI models. "说人话" (Shuo Ren Hua) offers a refinement skill designed for users of AI writing tools, enabling them to transform generic AI output into natural, human-like prose while strictly preserving factual accuracy and technical context. Its primary benefit is reducing the "AI tone" to allow for direct publishing of AI-assisted content, saving significant editing time and enhancing credibility.

How It Works

The core methodology prioritizes information fidelity over stylistic flair ("先保信息，再谈风格"). It operates by first identifying the text's context (e.g., chat, documentation, release notes) and then applying targeted refinements. Key features include "protected spans" that safeguard critical data like version numbers, commands, and factual attributions, and "Scene Packs" that tailor adjustments for specific publishing scenarios. The system assesses the severity of AI-like patterns and applies appropriate rewriting strategies, aiming for concrete actions, active voice, and context-specific tone, followed by a residual audit for any lingering unnaturalness.

Quick Start & Requirements

Installation involves cloning the repository (git clone https://github.com/MrGeDiao/shuorenhua.git) and navigating into the directory. Usage is demonstrated via command-line tools, such as invoking codex with the SKILL.md file as a system prompt: codex --system-prompt "$(cat SKILL.md)" "改写以下文本：...". Full functionality, including Scene Packs, requires the references/ directory alongside SKILL.md.

Highlighted Details

• Extensive Rule Coverage: Includes over 210 Chinese phrases, 96 English phrases, and 19 types of structural anti-patterns targeting common AI writing flaws.
• Robust Benchmarking: Features a 62-item benchmark set (35 "should fix," 27 "should not misfire") and 18 full-text real samples evaluated for naturalness, fidelity, and publishability.
• Scene Packs (v1.8.0): Advanced capabilities for refining specific content types like READMEs, release notes, forum posts, and issue replies, adapting the refinement strategy to the target publication venue.
• Information Preservation: Explicitly designed to protect and maintain crucial elements such as technical terms, commands, version numbers, error messages, and attribution.

Maintenance & Community

The project's latest release is v1.8.0, indicating active development. While the README does not detail specific community channels (like Discord/Slack) or a public roadmap, the MIT license suggests a permissive environment for contributions.

Licensing & Compatibility

The project is released under the MIT License, a permissive open-source license. This allows for broad compatibility, including use in commercial applications and integration within closed-source projects without significant restrictions.

Limitations & Caveats

This tool is not designed to bypass AI detection systems; its focus is solely on improving naturalness and reducing generic tone. It aims to reduce template-like output rather than perfectly emulate a specific individual's writing style. While effective, some subtle AI characteristics might persist, and the tool is primarily optimized for Chinese text refinement, with English support being secondary.