A Claude Cowork Skill
Strip the AI out of AI written copy. Keep everything else exactly as it is.
Humanize is a Claude Cowork skill that rewrites AI generated copy to remove every detectable AI writing pattern the inflated vocabulary, the em dashes, the fake profundity without touching your meaning, your length, or your intent.
It works through seven categories of AI tells on every pass:
- AI vocabulary: "pivotal," "tapestry," "underscore," "delve," "vibrant," and 50+ others
- Em dashes :replaced completely with commas, colons, or sentence breaks. No em dashes remain in the output.
- Puffery and fake significance: trailing significance claims, hollow legacy phrases, unsupported impact language
- "Not just X, but Y" patterns: collapsed into plain, direct statements
- Mechanical bold formatting: over-bolded key terms cleaned up to readable prose
- Chatbot sign-offs: "I hope this helps," "Let me know if..." deleted
- Knowledge-cutoff disclaimers: AI boilerplate hedges removed or rewritten as plain factual statements
After each pass, the skill delivers a plain-language change report covering exactly what was changed and why. It then asks whether you want further edits or to download the result as a Word doc (.docx), plain text (.txt), or Markdown (.md) file.
No slash commands. Once installed, activate it with natural language:
- "Humanize this"
- "Make this sound human"
- "Remove the AI tells"
- "This sounds like ChatGPT, fix it"
- "Clean up the AI writing"
Paste your text and the skill does the rest.
- Download or clone this repo
- Open Claude Cowork and go to Settings
- Select Capabilities, then Skills
- Click Upload Skill and select the
Humanizefolder (zipped) - Toggle the skill on and start a new conversation
Structural refactor to align with the Anthropic Agent Skill standard. No change to the rewrite logic; the skill produces the same output on the same input as v1.0.0.
- Moved
license: MITto a top-level key per standard - Added top-level
compatibilityfield, platform-agnostic - Added
metadata.homepagepointing to https://smbconsultants.ai - Extracted the AI vocabulary list to
references/ai-vocabulary.mdand wired it into SKILL.md - Added
VERSION.mdandREADME.md - Removed author/homepage block from SKILL.md body; now lives in
metadata - Expanded vocabulary list from ~37 entries to ~59, organized into five sections: high-priority replacements, corporate/consultancy tells, inspirational/aspirational tells, marketing/lifestyle tells, and inflation tells
- Added three new throat-clearing opener patterns: "In an era where...", "At the heart of...", "It's worth noting that..."
- Removed all em dashes from SKILL.md body (self-consistency fix the skill's own hard rule is em dash elimination)
- Added em dash check to the Step 3 final review pass
- Humanize rewrites AI generated copy to remove detectable AI writing patterns across seven categories: overused vocabulary, structural patterns, puffery and significance inflation, copula avoidance, formatting artifacts, collaborative chatbot language, and knowledge-cutoff disclaimers
- Em dashes are eliminated entirely
- The skill preserves the original meaning, length, and intent throughout
- After each pass, it delivers a plain-language change report and offers further edits or file download in Word, plain text, or Markdown
- Activated via natural language, no slash commands required
Part of The Samaritan Project
This tool was built for and extracted from The Samaritan Project: a build-in-public, self-hosted, multi-agent OSINT and intelligence research platform built on OpenClaw running on Parrot OS bare metal.
Support The Samaritan Project at buymeacoffee.com/thesamaritanproject
MIT-free to use, modify, and distribute.
Harold Mansfield, AI Consultant & Integration Specialst | Sec+ CySA+