vincentold/stop-slop-zh

Chinese written by AI no longer looks like AI-generated Chinese.

This is a set of prompt skills for models like Claude, GPT, Gemini, Wenxin, Tongyi, DeepSeek, to identify and eliminate typical traces of Chinese AI writing: clichés, parallelism, nominalized verbs, abstract subjects, memorable sentence endings, and total scores and eight essays.

Inspired by hardikpandya/stop-slop English projects, but not translation—the Chinese AI-flavored fingerprints are completely different from the English AI-flavored ones; the entire set of rules, vocabulary, sentence structures, and scoring is redone for Chinese.

---

What does this set of rules solve?

You’ve probably seen this kind of Chinese phrase:

In today’s era of rapid AI development, we are facing unprecedented opportunities and challenges. AI is not just a tool, but a way of thinking—a key force driving social progress. Let us embrace change with a more open attitude and create a better future together.

Every sentence is correct, but you can immediately tell it was written by AI. This skill aims to solve this problem of “every sentence is right, but together it forms AI.”

Project Structure

How to use it

Method 1: Claude Code (recommended)

Place the entire directory into the skills path that Claude Code can scan, or directly cd to this directory. The first dialogue tells Claude:

Please rewrite the following Chinese paragraph according to the rules of SKILL.md. After editing, score the five dimensions.

Claude automatically reads SKILL.md and references.

Method 2: Claude Projects / ChatGPT Projects

Upload both SKILL.md and references/*.md to the Project’s knowledge base. Then, in the system prompt or the first dialogue, write:

You play as the Chinese AI flavor removal assistant. All paraphrasing tasks must follow the 8 rules in SKILL.md and a 5-dimensional score. When ambiguous, prioritize looking up references/phrases.md and references/structures.md.

Method 3: One-time API call / Cherry Studio / Other Clients

Insert the full text of SKILL.md into the system prompt. references/*.md can be inserted as needed or combined into a long prompt.

Minimum available prompt template:

Method 4: Use as a manual proofreading checklist

It can be used even without connecting to LLMs. SKILL.md Section 4, “Pre-delivery Checklist,” consists of 9 manually checked items. After finishing the article, walk through the checklist.

Usage Recommendations

• Don’t check while writing. Finish what you want to say first, then rewrite it with this skill. Writing while looking up will lose your ideas.
• Starting from phrases.md categories 1 / 5 / 7 / 10: These four categories (opening clichés / degree adverbs / motivational endings / saturated value judgments) have the highest hit rates and fastest results.
• The hardest-hit area is the three-piece set layout.