Multi-Platform Prohibited Word Detection

Problem It Solves

Before publishing across platforms, operators and creators often have to check copy against different rules for WeChat Official Accounts, Xiaohongshu, and Douyin. A simple keyword scan can flag risky terms, but it does not explain how to replace them while preserving tone and layout. This skill is aimed at pre-publish review: it highlights problematic expressions in the copy, suggests context-aware replacements, and returns a revised version that is easier to publish after editing.

How It Works

The skill authenticates with REDFOX_API_KEY and calls a detection service built around official prohibited-word libraries. Core scripts include check_sensitive_words.py and extract_text.py. Input can be pasted text, a txt file, text extracted from an image, or the body of a web page; for URLs, the skill extracts page content and can handle some JS-rendered and SSR pages.

Detection is platform-specific: it defaults to WeChat Official Account rules, then switches to the corresponding library when the prompt mentions 'Xiaohongshu' or 'Douyin'. Copy over 3000 characters triggers a batching prompt, and copy over 10000 characters stops with a suggestion to split manually. Output is usually split into three parts: bold-marked hits, a replacement table, and a revised copy block, where suggested replacements are marked in bold italics for quick review.

Boundaries and Caveats

This skill focuses on text compliance checking. It does not analyze image style, color, or composition, and it does not support PDF or Word files. Image input is used only to extract text. For heavily protected, login-gated, or heavily dynamic web pages, manually copying the article body may be more reliable. The output is a compliance suggestion, not a substitute for legal review or final platform moderation; regulated industries should still verify claims against business scope and actual product effects.