360 Secure SkillHub
360 安全技能中心

Content System Humanizer Zh

该 Skill 检测中文正文中明显的 AI 写作痕迹,并对正文段落进行一次局部改写;在保留事实、数字、标题、结构与文章原意不变的前提下,生成改写后的副本并附一份 JSON 报告。约束是不编造事实或来源、不弱化鲜明观点、默认不改标题。

abigale-cyber
abigale-cyber
热度 64.49
最近更新 2026-05-11 17:40:59
分类:效率依赖:代码执行 / 读写文件应用场景:校对润色 / 文案创作 / 博客写作

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add content-system-humanizer-zh
Metadata
name
humanizer-zh
description
Remove obvious AI-writing traces from Chinese text in a constrained way. Use when Codex needs to reduce AI smell without changing facts, data, or the article's core argument.

Humanizer ZH

This repo-local skill is inspired by op7418/Humanizer-zh and blader/humanizer, but implemented for the content-system runtime.

Quick Start

.venv/bin/python -m skill_runtime.cli run-skill humanizer-zh \
  --input content-production/drafts/harness-engineering-一人公司-article.md

What It Does

  • Detects obvious AI-writing traces in Chinese prose
  • Applies a surgical rewrite pass on body paragraphs
  • Preserves facts, numbers, headings, structure, and article intent
  • Writes a humanized copy plus a JSON report

Output

  • content-production/drafts/<slug>-humanized.md
  • content-production/drafts/<slug>-humanizer-report.json

Constraints

  • Do not invent new facts or sources
  • Do not flatten strong viewpoints into neutral filler
  • Do not rewrite headings or titles by default
  • Do not pretend the text comes from personal experience if it does not