Forbidden Word Detection and Optimization

The Specific Problem: Beyond Simple Word Matching to Systematic Compliance Risk

Before publishing copy, content creators and social media operators often face the vague but high-stakes judgment: "Is this ready to go live?" The problem extends far beyond a few sensitive words. Simple dictionary-based checks lead to false positives (e.g., flagging "the best apple") or miss contextual violations like homophones and intentional misspellings. Deeper compliance risks exist across three layers: legal mandates (e.g., absolute terms under China's Advertising Law), platform-specific rules (e.g., content moderation and throttling policies on Xiaohongshu, Douyin, or WeChat), and content quality factors (e.g., missing sources leading to lower algorithmic distribution). These risks are interconnected, but traditional methods often only treat symptoms without a structured, traceable audit process.

Core Capability: Three-Layer Structural Scanning and a Structured Report

The core value of this skill is providing a "pre-publication compliance checkup." Its workflow involves several key steps:

1. Sequential Scanning Across Three Risk Layers: Instead of flat matching, it processes risk in a prioritized hierarchy:

   • Regulatory Compliance Layer (High-Risk): Checks for legal red lines such as absolute terms, false advertising, and missing citations. Judgment is based on regulations like the Advertising Absolute Term Law Enforcement Guidelines, interpreting "same or similar meaning."
   • Platform Rules Layer (Medium-Risk): Scans against the specific community or commercial rules of the target channel (e.g., Xiaohongshu, Douyin), linking directly to platform-specific throttling and penalty mechanisms.
   • Content Quality Layer (Low-Risk): Evaluates factors affecting recommendation weight, such as source reliability and clarity of expression.

2. Context-Aware Judgment: The skill does not rely on a rigid word list. It analyzes context to differentiate between subjective personal opinions (e.g., "best-tasting") and prohibited claims, and incorporates a whitelist exclusion mechanism (e.g., 6 categories of legal exceptions) to reduce false positives.

3. Generating a Detailed, Three-Column Issue List: For each problem found, it produces a clear, actionable report with:

   • Original Sentence/Location: The specific text with the issue.
   • Violated Clause/Risk Level: The exact law, platform rule, or content guideline breached, and whether it is high, medium, or low risk.
   • Compliant Rewrite Suggestion: Provides specific directions or examples for revision. For high-risk issues, the server may provide a revised, ready-to-use draft.

4. Publication Verdict and Revision Tracking: Based on the risk count from the scan, it automatically concludes whether "revision required," "optimization recommended," or "ready to publish." The skill also supports recording revision traces, aligning with standard editorial review workflows.

Scope of Application and Key Considerations

This is a focused copy text audit tool, and its boundaries must be clear:

• What it does: Audits pre-publication text copy for compliance and risk, given a specified industry (e.g., cosmetics, food) and publishing channel.
• What it does not do:
  • It is not for copywriting or ghostwriting: It only provides compliance suggestions within the audit context, not original creation or stylistic rewriting.
  • It does not audit images or video content: Visual elements like watermarks, before-and-after medical photos, or unmarked AI-generated content are out of scope.
  • It does not diagnose account or store violations: It only analyzes a single piece of copy, not the historical violation status or penalty points of an account.
  • It does not monitor post-publication sentiment: It performs only a static, pre-publish check.

Key Inputs: For the most accurate report, it is best to provide the complete copy (title + body + hashtags), publishing channel, and industry. Industry information can be omitted (the check will proceed with general regulatory rules, but providing it enables detection of industry-specific jargon), while the publishing channel is a mandatory field to load the corresponding platform rule set.