Skip to content
← Tools / Skills
Skill · .md packageblader/10 trustVerified
Humanizer logo

Humanizer

Agent skill that strips AI tells from prose without changing meaning

Visit official site →

About

Humanizer is an installable skill that rewrites AI-sounding text so it reads like a person wrote it. It checks for 26 patterns — not-X-but-Y contrasts, one-line closers, forced triads, em-dash spray, sales language, chatbot residue — based on Wikipedia's "Signs of AI writing" guide, and it can match your voice from a writing sample. It answers to /humanizer in Claude Code, installs into Codex, Gemini CLI, Copilot, Cursor, and Windsurf, and shows its work: first rewrite, critique, final version. It never invents facts — missing details get flagged, not filled in. The repo carries 56k+ stars and 6 releases under MIT. Honest limits: it targets prose style, not factual accuracy; AI detectors may still flag its output.

What it can do

  • 26-pattern AI-tell detection and rewrite
  • Voice matching from a writing sample
  • First-rewrite, critique, final-version workflow
  • File-targeted prose editing (code and data untouched)
  • Works across Claude Code, Codex, Cursor, Gemini, Copilot

Setup

Run npx skills add blader/humanizer --global --agent '*' (or target one agent). In Claude Code, register the plugin marketplace and install humanizer@humanizer, then call /humanizer.

Prerequisites

Node.js for the Skills CLI; Claude Code 2.1.142+ for the plugin route.

Use cases

rewriting AI draftsediting docs and postsvoice matchingde-slopping agent output

Works with

Claude CodeCodexCursorgemini-cligithub-copilotWindsurf

Trust score — 9/10

/10 trustVerified

Scored 9/10 (verified tier, metrics-v1). Evidence: Live GitHub page read this session: 56,332 stars, MIT license, 97 commits, 6 releases; install commands copied verbatim from README.; 26-pattern system based on Wikipedia's Signs of AI writing; /humanizer invocation; voice matching; works across Claude Code, Codex, Cursor, Gemini, Copilot, Windsurf.; README's own blind-test claim (16/16) is the author's figure — labeled as such. Detectors-may-still-flag caveat is from the README itself..

★ 56,332 GitHub stars

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.