yuwen-cool/yuwen-publish-precheck

What it solves

Content creators often face a dilemma where engaging topics are flagged as violations by platforms like Douyin, Xiaohongshu, and WeChat Channels. Because platforms don't provide pre-publication warnings, creators often only find out about violations after their content is restricted or penalized. This project provides a way to check content for compliance before publishing to avoid these penalties while maintaining the impact of the content.

How it works

The tool uses a two-layer detection process: first, a fast surface-level scan to identify potential problematic words, followed by a semantic analysis where an AI evaluates the context to determine if the phrasing is actually a violation. It avoids simple keyword lists by understanding that the same word can be acceptable in a personal story but prohibited in a commercial advertisement.

Users can also build a personal local knowledge base in a data/ directory. By giving the AI feedback (e.g., "Remember that 'fat burning' is okay in my fitness content"), the tool learns and becomes more accurate for the specific user's niche over time. It runs locally with Python 3 and is designed to be integrated as a "skill" for AI agents like Cursor or Claude Code.

Who it’s for

Content creators, marketers, and social media managers who publish text-based content (scripts, articles, captions, titles) to Chinese social media platforms (Douyin, Xiaohongshu, WeChat Channels).

Highlights

• Context-Aware Review: Distinguishes between harmless mentions and actual violations based on context rather than just keyword matching.
• Evidence-Based Rules: Based on 72 official citations from the Advertising Law, Cyberspace Administration of China, and platform community guidelines.
• Local Data Privacy: All personal rules, archives, and content are stored locally in plain text and are never uploaded to third parties.
• Iterative Learning: Allows users to save their own experience and "lessons learned" from platform penalties to refine future checks.
• Anti-Circumvention: Explicitly refuses to provide "hacks" or homophones to bypass filters, focusing instead on legitimate compliant phrasing.