Humanizer (blader/humanizer)

Properties

category: anti-detector
maturity: established
approach: A Markdown-only agent skill (not a hosted web app) that any coding agent supporting the skills format (Claude Code, Codex, OpenCode, Cursor) loads and follows to identify and rewrite AI-sounding text, drawing on patterns drawn from Wikipedia's Signs of AI writing project. Runs inside the agent's own context with no server round trip, no upload, and no separate inference call.
access: open-source
owns: The most widely adopted open-source text-humanization tool found in this catalogue: an MIT-licensed agent skill that rewrites AI-sounding prose to read more naturally while trying not to invent facts it was not given.
distinctFrom: Unlike undetectable-ai and stealth-writer (paid, hosted rewriting web services that a user pastes text into), Humanizer is a free, open-source skill file consumed directly by a coding agent's own context, with no server-side product at all; the baseline catalogue previously had no open-source entry in the anti-detector category. A large ecosystem of forks and similarly-named projects exists (jooray/humanizer, 49 stars; Aboudjem/humanizer-skill, 247 stars; crabin/paper-humanizer-skill, 122 stars; brandonwise/humanizer and WhimseyAI/humanizer-skill, single digits to low tens) and were checked and rejected as redundant with this original, far larger repo.
source: blader/humanizer (GitHub)
notes: Quote is the GitHub API's own repository description field, confirmed verbatim via the API. Created 2026-01-18, 50,178 stars and still actively pushed as of 2026-09-06 per the GitHub API, so it predates this lens's 2026-09-04 window by creation date but is included as a major untracked gap-fill given its scale; not a confirmed brand-new in-window launch. Do not cite a specific pattern count (sources disagree: a third-party blog says 35, a direct README fetch returned 25, and the unrelated Aboudjem/humanizer-skill repo says 55) or an accuracy figure; none is independently confirmed. The coddykit.com blog posts covering this repo report star counts (41,000-42,893) that contradict the GitHub API's actual 50,178 and were not used as a source for that reason.

Provenance

Source: https://github.com/blader/humanizer

Properties

category

anti-detector

maturity

established

approach

A Markdown-only agent skill (not a hosted web app) that any coding agent supporting the skills format (Claude Code, Codex, OpenCode, Cursor) loads and follows to identify and rewrite AI-sounding text, drawing on patterns drawn from Wikipedia's Signs of AI writing project. Runs inside the agent's own context with no server round trip, no upload, and no separate inference call.

access

open-source

owns

The most widely adopted open-source text-humanization tool found in this catalogue: an MIT-licensed agent skill that rewrites AI-sounding prose to read more naturally while trying not to invent facts it was not given.

distinctFrom

Unlike undetectable-ai and stealth-writer (paid, hosted rewriting web services that a user pastes text into), Humanizer is a free, open-source skill file consumed directly by a coding agent's own context, with no server-side product at all; the baseline catalogue previously had no open-source entry in the anti-detector category. A large ecosystem of forks and similarly-named projects exists (jooray/humanizer, 49 stars; Aboudjem/humanizer-skill, 247 stars; crabin/paper-humanizer-skill, 122 stars; brandonwise/humanizer and WhimseyAI/humanizer-skill, single digits to low tens) and were checked and rejected as redundant with this original, far larger repo.

source

blader/humanizer (GitHub)

notes

Quote is the GitHub API's own repository description field, confirmed verbatim via the API. Created 2026-01-18, 50,178 stars and still actively pushed as of 2026-09-06 per the GitHub API, so it predates this lens's 2026-09-04 window by creation date but is included as a major untracked gap-fill given its scale; not a confirmed brand-new in-window launch. Do not cite a specific pattern count (sources disagree: a third-party blog says 35, a direct README fetch returned 25, and the unrelated Aboudjem/humanizer-skill repo says 55) or an accuracy figure; none is independently confirmed. The coddykit.com blog posts covering this repo report star counts (41,000-42,893) that contradict the GitHub API's actual 50,178 and were not used as a source for that reason.