humanize

A Claude Code plugin that strips robot-sounding language from user-facing content.

What it does

Detects and removes 15 categories of machine-generated writing patterns:

• Em-dashes used for dramatic effect
• Corporate jargon ("leverage", "robust", "seamless", "ecosystem")
• Passive voice ("is processed by", "was created")
• Hedge phrases ("might potentially", "could perhaps")
• Buzzword clusters ("optimized, scalable, future-proof")
• Stiff constructions ("It is important to note that...")
• Redundancy ("advance planning", "past history")
• Over-formal transitions ("Furthermore", "Moreover")
• Plural first person in solo contexts ("we" when "I" is correct)
• LLM attribution (removes references to the model as author)
• And more

Each pattern has a confidence score. High-confidence patterns are auto-fixed, medium-confidence patterns generate suggestions, and low-confidence patterns are flagged for human review.

Install

claude plugin add lukeslp/skill-humanize

Usage

In Claude Code, invoke the slash command:

/humanize:humanize

Or ask Claude Code to humanize your content:

Humanize the README before I publish this project

Safety

• Never modifies code blocks
• Never changes URLs or citations
• Preserves technical specifications
• Creates git checkpoints before making changes
• Shows diff previews for all changes

License

MIT

Author

Luke Steuber (luke@dr.eamer.dev)