Writing Humanizer

Remove AI writing patterns from text, making it sound more natural and human. Focused on Traditional Chinese (Taiwan).

One skill, seven hosts. The same skills/writing-humanizer skill runs in Claude Code, OpenAI Codex, Cursor, Google Antigravity CLI, OpenCode, Kimi CLI, and pi — only the per-host manifest differs.

Installation

Claude Code

Via marketplace (recommended):

/plugin marketplace add shyuan/shyuan-marketplace
/plugin install writing-humanizer@shyuan-marketplace

Or direct:

/plugin install shyuan/writing-humanizer

OpenCode

Add to the plugin array in your opencode.json (global or project-level), then restart OpenCode:

{
  "plugin": ["writing-humanizer@git+https://github.com/shyuan/writing-humanizer.git"]
}

See .opencode/INSTALL.md for version pinning and troubleshooting.

Codex / Cursor / Antigravity CLI / Kimi CLI

Each of these hosts reads its own manifest bundled in this repo. Install with the host's own plugin/marketplace command pointing at this repo, or clone it into the host's plugins location and let the host discover the manifest:

git clone https://github.com/shyuan/writing-humanizer.git

Host	Manifest the host reads
OpenAI Codex	`.codex-plugin/plugin.json`
Cursor	`.cursor-plugin/plugin.json`
Google Antigravity CLI	`plugin.json` (repo root)
Kimi CLI	`.kimi-plugin/plugin.json`

pi

pi loads the skill through the "pi" field in package.json (extension at .pi/extensions/writing-humanizer.ts). Install the package into your pi setup so pi discovers the bundled skills/.

What You Get

Skill: `writing-humanizer`

A comprehensive skill that detects and rewrites 31 categories of AI writing patterns:

Content Patterns (1-6)

Significance inflation, notability name-dropping, superficial analysis
Promotional language, vague attributions, formulaic "challenges & outlook"

Language Patterns (7-12)

AI vocabulary overuse, copula avoidance, negative parallelisms
Rule of three, synonym cycling, false ranges

Style Patterns (13-17)

Em dash overuse, boldface overuse, inline-header lists
Emoji decoration, quotation mark issues

Communication Patterns (18-24)

Chatbot artifacts, knowledge-cutoff disclaimers, sycophantic tone
Filler phrases, excessive hedging, generic conclusions, throat-clearing openers

Structure & Rhetoric Patterns (25-31) — Chinese-essay-specific

Outline-as-essay, bold 4-char-label lists, significance-stamping endings
In-sentence keyword bolding, meta-discourse/roadmap announcements, sublimating moralizing conclusions
Reifying an abstract process into "a line/route" and pointing back to it repeatedly

Key Features

AI vocabulary watchlist with suggested replacements for Traditional Chinese (Taiwan)
Common AI sentence templates specific to Chinese writing
Two-pass rewriting process with self-audit
Quality scoring system (50 points across 5 dimensions)

Quick Start

After installation, invoke the skill (slash-command syntax varies by host — this is the Claude Code form):

/writing-humanizer:writing-humanizer <paste your text here>

Or simply ask your agent to humanize text — the skill is discovered from its description and loaded on demand.

References

License

MIT