lov-anti-wechat-ai-check

anti-wechat-ai-check — WeChat Official Account AI trace detection and humanizing polish

Detects AI-generated traces in articles (template phrases, piled-up transition words, repetitive sentence patterns, etc.), gives a risk score and revision suggestions, and can output a humanized polished version. Based on the WeChat Official Account Platform Operation Norms clause 3.27 (non-human automated creation behavior) detection logic.

When to Use

• The user is preparing to publish an AI-assisted writing article to a WeChat Official Account
• The user wants to check whether an article has obvious AI traces
• The user wants to rewrite an AI-generated draft into a more natural human style

Workflow (MANDATORY)

You MUST follow these steps in order:

Step 1: Get the article

Determine the input source:

• If user provides a file path → read the file
• If user pastes text in the conversation → save to a temp file or use --text

Step 2: Run analysis

python skills/lov-anti-wechat-ai-check/scripts/analyze.py \
  --input <path> --format json

Or with inline text:

python skills/lov-anti-wechat-ai-check/scripts/analyze.py \
  --text "文章内容" --format json

Step 3: Present findings

Show the user:

1. Risk score (0-100) and risk level (LOW / MEDIUM / HIGH)
2. Template phrases found — list each one with its location
3. Structure issues — transition word density, paragraph uniformity, etc.
4. Sentence issues — length uniformity, repeated starters, excessive "的"

Step 4: Ask the user

IMPORTANT: Use AskUserQuestion to ask what to do next:

Option	Description
View report only	The user revises it themselves, skill ends
Give revision suggestions	List specific revision suggestions for each issue, without changing the original text
Directly output the revised version	Output the complete article after humanizing polish

Step 5: Humanize (if requested)

When rewriting, follow these humanization rules:

5a. Eliminate template phrases

• Delete or replace every template phrase marked in the report
• "随着科技的不断发展" → say the specific thing directly ("去年 ChatGPT 发布后...")
• "综上所述" → delete it, or replace it with a colloquial ending

5b. Reduce transition word density

• Goal: transition word density < 15%
• Delete unnecessary "首先/其次/此外/另外"
• Replace generalized connectives with specific logical relationships

5c. Break sentence uniformity

• Deliberately create alternation of long and short sentences: short sentences < 15 characters, long sentences > 40 characters
• Add colloquial expressions, rhetorical questions, exclamatory sentences
• Occasionally use incomplete sentences or elliptical sentences

5d. Break paragraph uniformity

• Some paragraphs have only one or two sentences, while some paragraphs can be very long
• Avoid every paragraph being the three-part pattern of "argument + evidence + summary"

5e. Add human flavor

• Add personal experiences, specific cases, numerical details
• Use colloquial expressions ("说白了"、"讲真"、"你想想")
• Appropriately use non-standard but natural expressions
• Reduce use of the character "的" (goal < 5%)

5f. Preserve the original meaning

• Core viewpoints and information must not be lost
• Keep technical terms, do not over-colloquialize
• Maintain the original text's stance and attitude

Step 6: Output

Output the humanized article as markdown. If the input was a file, also offer to write the result back to a file (with -humanized suffix).

CLI Reference

Argument	Default	Description
--input, -i	—	Input file path (.md, .txt)
--text, -t	—	Inline text to analyze
--format, -f	text	Output format: text or json

Dependencies

No external dependencies — stdlib only.

Runtime context (shared)

Before running, read this Skill package's skill.yaml, with the host providing the skill-runtime/v1 context. The field resolution order is: current request, project context, personal Preferences, brand Profile, general defaults.

• Use only fields declared in the Manifest; Profile stores public brand facts, Preferences stores personal work preferences.
• required: true When a field is missing, ask the user one focused question according to the Manifest's question configuration; save the answer only after the user explicitly agrees.
• Errors provide a copyable context_id, field path and source; diagnostic content avoids secrets, full private paths and raw configuration.

General feedback loop

When the user raises revision opinions during a Skill-driven task, before continuing the current artifact you must execute:

1. First determine whether the opinion is task-specific(this time only) or reusable(reusable across tasks).
2. task-specific Only modify the current task, do not change the Skill.
3. reusable First determine the scope: for domain rules, update the corresponding canonical Skill first; for rules that apply to all Skills, update the shared specification first.
4. After completing the rule update, versioning, lint, and distribution verification, then apply the changes to the current task.
5. reusable The modification will invalidate previous "confirm," "continue," and "go ahead"; after completing the current artifact modification and read-back, you must stop and wait for the user's next instruction, and do not automatically proceed to publishing, committing, or other external writes.