humanize-writing

A Claude Code skill that rewrites AI-generated content to sound like a human wrote it.

What it does

Detects and fixes common AI writing patterns through an 8-pass editing process:

Structure tells -- formulaic sections, identical list lengths, tidy takeaways
Significance inflation / promotional language -- "pivotal moment," "nestled," "vibrant"
AI vocabulary -- "delve," "landscape," "leverage," "tapestry," and dozens more
Grammar-level patterns -- copula avoidance, -ing phrases, synonym cycling, false ranges
Rhythm and style -- metronomic cadence, em dash overuse, boldface, emoji, curly quotes
Hedging, filler, and vague attributions -- "It's worth noting," chatbot artifacts, sycophancy
Connective tissue -- "Moreover," "Furthermore," "Additionally" on repeat
Human texture and soul -- adding opinions, lived experience, personality

See SKILL.md for the full editing process and references/ai-tells.md for the complete detection checklist.

npx skills add jpeggdev/humanize-writing

npm install -g @jpeggdev/humanize-writing

The postinstall script copies the skill files to ~/.claude/skills/humanize-writing/ (and equivalent directories for Cursor and Windsurf).

Once installed, the skill activates automatically in Claude Code when you say things like:

This skill combines two approaches:

The pass-based editing workflow and ai-tells reference from humanize-writing
The 24 Wikipedia-sourced patterns and "soul" philosophy from humanizer by blader

Primary reference: Wikipedia: Signs of AI writing, maintained by WikiProject AI Cleanup.

2.0.0 - Combined humanize-writing and humanizer-blader into a single skill. Added passes for significance inflation, grammar-level patterns, and soul/personality. Expanded ai-tells.md with additional patterns.
1.0.1 - npm packaging fixes
1.0.0 - Initial release

npm uninstall -g @jpeggdev/humanize-writing

This removes the skill files from all agent skill directories.

MIT