Douyin Prohibited Words Detection

Problem

Before publishing Douyin content, creators often hit traffic restrictions or takedowns caused by prohibited terms, superlative claims, or restricted promotional language. Manually checking scripts is slow and easy to miss, especially when risk words depend on context. @user_ebee0fcc/dy-prohibited-word turns that compliance check into a clear workflow: input a script, product copy, livestream talk track, or page text, then get risk markers, replacement suggestions, and a publish-ready version.

Workflow and Core Capabilities

The skill targets the Douyin platform and accepts several input types:
- Plain text: paste copy directly for length checking and detection.
- Documents: TXT, DOC, and DOCX files can be extracted and checked.
- Web pages or images: web pages are parsed for text content, while images are used only for text extraction and do not analyze visual information.

After detection, it outputs three fixed sections:
1. Prohibited-word results: number and types of matched terms.
2. Replacement suggestions: prohibited word, suggested replacement, and reason.
3. Optimized copy: a full revised version with replacements highlighted for easy reuse.

It also handles practical edge cases: content over 3000 characters triggers batching confirmation, 3000~10000 characters can be batched and merged after user confirmation, and English substring false positives are filtered to avoid misjudging ordinary English words. Dynamic pages prefer headless-browser rendering, with static extraction as a fallback.

Boundaries and Notes

The skill does not replace platform moderation and depends on an external detection service. Copy is sent over HTTPS to the service and is not stored locally; an API key must be configured before use. PDF files are not supported, and images are not analyzed for style, layout, people, or other non-text details. For content over 10000 characters, split it manually before detection.