AI analysis grounded in the code graph — computed facts, not vibes · 2026-09-08T03:03:19Z
Humanizer is a plugin/skill for AI coding and chat agents (Claude Code, Claude Desktop, and any agent supporting the "skills" convention) that rewrites AI-generated-sounding text into more natural prose. It is implemented as a Markdown specification (SKILL.md) rather than executable application code — the agent itself does the rewriting by following the patterns and instructions defined in that file, based on Wikipedia's "Signs of AI writing" list (35 documented patterns). The only Python present is tooling in scripts/ (build-skill-zip.py, validate-package.py) used to package and validate the skill for distribution, not to perform the humanizing itself.
The growth (1,208 stars in a single day, 41,610 total) tracks with a fast release cadence — four releases between v2.9.0 and v2.11.1 in under a month — and steady content iteration visible in commit titles (new vocabulary entries, broadened pattern coverage, plain-language rewrites of the skill itself). The README's install path via npx skills add and a Claude Code plugin marketplace listing lowers the adoption barrier considerably, which plausibly explains rapid uptake once discovered, though the graph facts alone (three symbols, one module) don't account for the scale of star growth — that's better explained by the README's install mechanics and topical timing than by anything in the codebase.
What changed recently, how it's actually built (from the code graph), and whether you should care. Free account — no card, no spam.