humanify

An agent skill for Claude Code, Codex, Antigravity, and Gemini CLI that detects and rewrites the linguistic patterns that make AI-drafted prose read as machine-generated - banned clichés, AI-favored vocabulary, copula substitution, participle-ending sentences, elegant variation, and more. Works on any long-form text: articles, posts, emails, docs.

Works with: Claude Code · Codex · Antigravity · Gemini CLI

Highlights

• Banned phrase list - clichéd pivots, weak contrast constructions, significance inflation, throat-clearing, faux-revelation transitions, and more
• AI vocabulary scan - flags words that appear at significantly higher rates in AI-generated text (vibrant, crucial, delve, tapestry, nuanced, transformative...) with a ranked priority tier
• Structural checks - copula substitution, participle endings, elegant variation chains, list-length uniformity, fragmented headers, punctuation personality
• Non-destructive dry run - --dry-run reports findings without editing and gives a findings-per-1,000-words density figure (with rough bands) so you can gauge how "AI" a draft sounds before committing to a revision
• Voice calibration - --voice <file> reads a sample of the author's own prior writing and suppresses house-style checks the sample shows are deliberate
• Protected content - quotes, citations, code, URLs, and proper nouns are flagged but never edited

Getting Started

Installation

humanify is a plain-markdown skill (the open Agent Skills SKILL.md format) with no dependencies or executable code, so it installs the same way on every supported agent.

Claude Code

From the mackayi marketplace (recommended), run inside a Claude Code session:

/plugin marketplace add keithmackay/mackayi
/plugin install humanify@mackayi

Or manually:

git clone https://github.com/keithmackay/humanify.git

# Global install (available in all projects)
ln -s "$(pwd)/humanify" ~/.claude/skills/humanify

# Project-local install (available only in one project)
ln -s "$(pwd)/humanify" /path/to/your/project/.claude/skills/humanify

Invoke with /humanify.

Codex

Clone the repo, then register it in your Codex plugin marketplace.

~/.agents/plugins/marketplace.json (create if absent):

{
  "name": "personal",
  "interface": { "displayName": "Personal Plugins" },
  "plugins": [
    {
      "name": "humanify",
      "source": { "source": "local", "path": "/path/to/humanify/" },
      "policy": { "installation": "AVAILABLE", "authentication": "ON_INSTALL" },
      "category": "Productivity"
    }
  ]
}

The plugin manifest is .codex-plugin/plugin.json; the skill itself is at skills/humanify/.

Antigravity

git clone https://github.com/keithmackay/humanify.git

# Global install (all workspaces)
cp -r humanify ~/.gemini/antigravity/skills/humanify

# Workspace install (current project only)
cp -r humanify /path/to/your/project/.agents/skills/humanify

Skills are auto-discovered. You can also mention humanify by name to force activation.

Gemini CLI

gemini extensions install https://github.com/keithmackay/humanify

To update: gemini extensions update humanify. The skill loads through GEMINI.md after installation.

Usage

The examples use the /humanify form. On Gemini CLI and Antigravity, ask the agent to run humanify (or mention it by name) and pass the same flags.

/humanify                 # check and rewrite the target text
/humanify --dry-run       # report findings only, no edits
/humanify --voice <file>  # calibrate house-style checks against a writing sample
/humanify --version       # prints the installed version and checks for updates
/humanify --help          # reads and displays help.md verbatim, takes no other action

help.md is the authoritative argument reference; SKILL.md is the source of truth for behavior - the full list of banned phrases, AI vocabulary tiers, and structural checks. Report templates live in references/report-format.md and the second-tier vocabulary list in references/tells.md.

Compatibility

Feature	Claude Code	Codex	Antigravity	Gemini CLI
Core skill (checks, edits, reports)	✅	✅	✅	✅
--dry-run, --voice, --version, --help	✅	✅	✅	✅
references/ sub-documents	✅	✅	✅	✅
Marketplace install	✅	✅	❌	❌
Install from GitHub URL	❌	❌	❌	✅

Legend: ✅ Supported · ❌ Not supported

humanify uses no platform-specific frontmatter, hooks, or subagents, so the same skill content runs unchanged on all four agents. The Antigravity copy is the root SKILL.md.

References

• Claude Code Skills: https://code.claude.com/docs/en/skills
• Codex Plugins: https://developers.openai.com/codex/plugins/build
• Antigravity Skills: https://antigravity.google/docs/skills
• Gemini CLI Extensions: https://github.com/google-gemini/gemini-cli/blob/main/docs/extension.md
• Agent Skills open standard: https://agentskills.io/home

About

Version 1.1.0 · MIT license
Source:  https://github.com/keithmackay/humanify
Author:  Keith MacKay <technofacto@gmail.com>
Update:  git -C ~/.claude/skills/humanify pull   (or: /plugin marketplace update mackayi; gemini extensions update humanify)
Issues:  https://github.com/keithmackay/humanify/issues

Contributing

Pull requests are welcome - fork the repo, make your change, and open a PR describing what it does and why.

Changelog

See CHANGELOG.md for release history.

License

MIT