natural-chinese-copy-core

This repository is a writing skill for AI coding assistants (specifically Codex and Claude) that helps produce natural-sounding Chinese copy. The project name translates roughly as "Chinese is the real deal," and the skill is built around a specific problem: AI-generated Chinese text often sounds stiff, machine-translated, or built on English sentence structure rather than natural Chinese phrasing.

The skill introduces a multi-step process before delivering any finished text. First it creates what the README calls a Chinese fact draft: a raw summary of the source material written directly in Chinese rather than translated from an English framing. Then it checks each key sentence for whether the subject, verb, and object are correctly matched, and whether the verb describes something the subject can actually do. Sentences that match a known list of bad patterns are discarded entirely and rewritten from the fact draft rather than patched in place.

After the automated checks pass, there is still a manual review stage for checking whether the output is appropriate for its intended audience, channel, and publication context. The project explicitly separates passing the automated validators from being ready to publish.

The repository also includes a feedback loop: when a user points out a bad sentence, the example goes into a sample file, gets reviewed, and may eventually become a new rule or validator. Single feedback cases do not automatically become hard rules, which protects the system from overreacting to one-off preferences.

The project ships as a directory of skill files, reference documents, and Node.js validation scripts. It is compatible with both Codex and Claude and requires no external services or accounts to use. The README is written entirely in Chinese, reflecting its intended audience of Chinese-language copywriters and content creators.