Humanizer (Chinese) (Claude Skill) — Install & Live Demo

Why use it

Key features

Targeted at Chinese AI-tells, not a generic English humanizer
Multiple voice profiles: 公众号 / 小红书 / blog / Twitter / 学术
Detects + removes overused phrases (其实, 总的来说, 综上所述...)
Sentence-rhythm variation — breaks the predictable cadence
Preserves technical accuracy (won't soften proper nouns, code, numbers)
Pure markdown skill — no external dependencies

Live Demo

What it looks like in practice

ready

Install

Pick your client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "humanizer-zh-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/op7418/Humanizer-zh",
        "~/.claude/skills/humanizer-zh"
      ],
      "_inferred": true
    }
  }
}

Open Claude Desktop → Settings → Developer → Edit Config. Restart after saving.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "humanizer-zh-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/op7418/Humanizer-zh",
        "~/.claude/skills/humanizer-zh"
      ],
      "_inferred": true
    }
  }
}

Cursor uses the same mcpServers schema as Claude Desktop. Project config wins over global.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "humanizer-zh-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/op7418/Humanizer-zh",
        "~/.claude/skills/humanizer-zh"
      ],
      "_inferred": true
    }
  }
}

Click the MCP Servers icon in the Cline sidebar, then "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "humanizer-zh-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/op7418/Humanizer-zh",
        "~/.claude/skills/humanizer-zh"
      ],
      "_inferred": true
    }
  }
}

Same shape as Claude Desktop. Restart Windsurf to pick up changes.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "humanizer-zh-skill",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/op7418/Humanizer-zh",
        "~/.claude/skills/humanizer-zh"
      ]
    }
  ]
}

Continue uses an array of server objects rather than a map.

~/.config/zed/settings.json

{
  "context_servers": {
    "humanizer-zh-skill": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/op7418/Humanizer-zh",
          "~/.claude/skills/humanizer-zh"
        ]
      }
    }
  }
}

Add to context_servers. Zed hot-reloads on save.

claude mcp add humanizer-zh-skill -- git clone https://github.com/op7418/Humanizer-zh ~/.claude/skills/humanizer-zh

One-liner. Verify with claude mcp list. Remove with claude mcp remove.

Use Cases

Real-world ways to use Humanizer (Chinese)

Rewrite an AI-generated blog post in 公众号 voice

👤 中文 content creators ⏱ ~25 min beginner

When to use: Claude wrote a tech post and it reads like a translated whitepaper.

Prerequisites

Skill installed — git clone https://github.com/op7418/Humanizer-zh ~/.claude/skills/humanizer-zh
Source draft — Markdown file with the AI-generated post

Flow

Identify AI tells

Use humanizer-zh. Pass 1: scan /drafts/post.md. List the top 10 sentences that scream AI — quote them.✓ Copied

→ Concrete sentence-level diagnoses, not generic feedback
Rewrite

Now rewrite the whole post in 公众号 voice. Keep technical accuracy; change rhythm and register.✓ Copied

→ Output reads like a person wrote it; no '众所周知' / '总而言之' patterns
Diff review

Show the diff section by section. I'll flag any over-translation.✓ Copied

→ Clear before/after for each paragraph

Outcome: Post reads natural; passes a reader's BS detector.

Pitfalls

Skill softens technical claims (e.g. 'is' → 'might be') — Add to the prompt: 不要软化技术结论, only change register

Combine with: filesystem

Convert a long article into 小红书笔记 format

👤 小红书 creators ⏱ ~10 min beginner

When to use: You have a 1500-word article and need a 小红书 post that follows the platform's conventions.

Flow

Apply 小红书 profile

Humanizer-zh: 小红书 mode. Reduce to 600 字, add emoji where natural, break into short paragraphs, add a hooky 标题.✓ Copied

→ Output matches platform tone; no obvious AI emoji-stuffing

Outcome: Ready-to-post 小红书笔记.

Pitfalls

Emoji over-stuffing — Cap at 1 emoji per ~80 字

Humanize a tutorial without breaking code

👤 Tech writers ⏱ ~15 min intermediate

When to use: Tutorials are sensitive — you want voice change but not corrupted code blocks.

Flow

Mark code blocks

Humanizer-zh: rewrite /tutorials/python.md. Treat all ``` blocks as untouchable. Only rewrite prose.✓ Copied

→ Code untouched; prose rewritten

Outcome: Tutorial reads natural while code stays correct.

Pitfalls

Inline code formatted with … also gets rewritten — Explicit prompt: protect inline code too

Combinations

Pair with other MCPs for X10 leverage

humanizer-zh-skill + filesystem

Batch process all drafts in a folder

Humanizer-zh on every .md under /drafts/. Save as /drafts-final/.✓ Copied

humanizer-zh-skill + x-article-publisher-skill

Humanize then publish

Humanize the post, then publish to X via x-article-publisher.✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call
diagnose	text	Pass 1 — understand what to fix
rewrite	text, voice_profile?	Pass 2 — apply the change
diff_review	before, after	Pass 3 — verify nothing was over-changed

Cost & Limits

What this costs to run

API quota: N/A
Tokens per call: Roughly 2x source length (read + rewrite)
Monetary: Free (the skill itself); your model tokens apply
Tip: Diagnose first on the whole doc, but rewrite section-by-section to keep context tight

Security

Permissions, secrets, blast radius

Minimum scopes: filesystem-read filesystem-write

Credential storage: None

Data egress: None (skill is local)

Always preserve original drafts; the skill rewrites can shift nuance

Troubleshooting

Common errors and fixes

Output sounds equally robotic

Specify voice profile; default mode is conservative

Code blocks corrupted

Add explicit 'protect ``` blocks' to prompt

Length drift (output much shorter)

Specify target length explicitly

Alternatives

Humanizer (Chinese) vs others

Alternative	When to use it instead	Tradeoff
Original Humanizer (English)	Source is English	Generic; less tuned for Chinese AI tells
Manual editing	Stakes are very high (e.g. published whitepaper)	Slow; doesn't scale

More

Resources

📖 Read the official README on GitHub

🐙 Browse open issues

🔍 Browse all 400+ MCP servers and Skills