WeChat Official Account Prohibited Word Detection

Problem to Solve

Official-account posts, H5 landing pages, and poster copy often get flagged or removed for advertising-law superlatives, medical aesthetics claims, financial yield promises, or exaggerated education claims. Manual word-by-word review is slow, and simple synonym replacement can make the copy sound unnatural. This skill turns pre-publish compliance review into a structured workflow: identify risky phrases, suggest replacements, and produce a compliant final version.

Core Capabilities and Workflow

• Multiple input modes: paste text, upload TXT/DOC/DOCX files, upload screenshots for text extraction, or provide a web URL for body-content extraction.
• Detection and highlighting: calls a remote detection service covering 10+ categories, including advertising law, medical aesthetics, financial risk, and education; hits are shown in bold, with a summary of types and counts.
• Replacement and final draft: each risky term gets a context-aware replacement suggestion and rationale, then a full revised copy is generated with changed spots marked for review.
• Engineering handling: content over 3000 characters prompts batch detection at natural breaks and merges results; English substring false positives are filtered; retries handle 5xx, timeout, or connection failures.

Boundaries and Notes

It checks text compliance only and does not analyze visual design. PDF is not supported; convert it to text or an image. Keep single checks under 3000 characters when possible, and split content over 10000 characters manually. Some login-gated or anti-scraping pages may fail extraction, so copy the text and paste it. It runs as a Python script and uses requests, python-docx, beautifulsoup4, and playwright to handle different input sources.