work-humanizer: an Agent Skill that removes AI tone without changing what your text commits to

PawelHuryn · x · 2026-10-09

Developer Pawel Huryn open-sourced work-humanizer, an Agent Skill that strips AI tone from workplace writing without altering what the text says or commits to. Install with npx skills add phuryn/work-humanizer, then run "Use work-humanizer on this: <paste text>".

Why it exists: generic humanizer prompts silently change meaning. In his tests, a plain "rewrite this to sound human" prompt told a customer their £45 refund had already gone through when the original only said "we're refunding". Existing tools invent facts, attach conditions to offers, or assign blame — potentially with legal consequences.

Evidence: the repo ships blind-judge scores, hand-checked meaning-change logs, and Pangram AI-detection results. Target use cases: sales emails, marketing copy, support replies, HR notes, legal letters — any work writing where exact words are commitments. MIT licensed.

Original post →

More from coding & agent

coding & agent channel →