yuwen-cool/yuwen-publish-precheck

This project is an AI skill, meant to be installed into tools like Claude Code, Cursor, or Codex, that checks a piece of content before you publish it on Chinese social platforms such as Douyin, Xiaohongshu, or WeChat Channels. The core problem it addresses is that creators often do not find out a post violates platform rules until after it is published and already penalized, since platforms rarely warn creators in advance. Rather than relying on a simple banned word list, the tool works in two steps. First a script quickly scans the draft for candidate phrases that might be risky. Then the AI evaluates each candidate in context, since the same words can be fine or against the rules depending on how they are used, such as sharing a personal experience versus using the same phrase to sell a product. If it finds a real problem, it suggests a specific rewritten sentence that keeps the original tone and hook while removing the violation, then rechecks the revised text automatically. Its rule base is built from 38 real world scripts that platforms actually approved, so it calibrates what counts as acceptable rather than overreacting. Every rule it applies is tied to a real, citable source, such as advertising law or official platform community guidelines, and the README states there are 72 referenceable citations behind its judgments. It explicitly refuses to teach ways to sneak past moderation, such as replacing risky words with homophones or symbols, stating that trading a small risk for a bigger one is not worth it. The tool also builds a personal, local rule library over time. A user can tell it in plain language to watch for a certain phrase going forward, or to stop flagging a word that is normal in their specific niche, and this personal data is stored only in a local folder on the user's own computer, never uploaded anywhere. It runs entirely locally using only Python, with no other dependencies, and it explicitly states it does not guarantee any piece of content will actually be approved by a platform, since final moderation decisions depend on factors the tool cannot see.