Open sourcePython · ≈62k stars · MIT

Humanizer

Name: Humanizer
Author: blader

An agent skill that strips signs of AI-generated writing from text.

blader/humanizer

Maintained by blader

Vertical Agents

Humanizer is an agent skill that removes the tell-tale signs of AI-generated writing from text. Install it as a Claude Code plugin (/humanizer:humanizer), or with npx skills for Codex or any other agent.

View on GitHub Homepage

Pros & cons

Each point cites its sources and says what kind of evidence backs it. A second reviewer tried to refute every claim before it was kept.

Pros

DocsBuilt on Wikipedia's "Signs of AI writing" guide, it checks 26 named patterns.[1]
User reportsIn a user-run blind study, the rewrite won 16 of 16 quality comparisons against the original AI text.[1]
DocsInstalls as a Claude Code plugin or through `npx skills` for Codex and other agents.[1]
MetadataMIT-licensed, about 62k stars, and 3 open issues.[1]

Cons

DocsIt does not beat AI detectors: the README says detectors still flag most of its output, and the same blind study found no change in detection.[1][2]
User reportsIt is a prompt-only skill whose results depend on the host model, and some users report it "doesn't do anything" visible.[1]
User reportsEarly releases were rejected by Claude.ai's "Add a Skill" upload, and the Claude Code plugin now needs version 2.1.142 or newer.[1][2][3]
User reportsIt can over-correct text that is already human-written; users asked for a density pre-check to avoid that.[1]

Best for: Writers and marketers cleaning up AI-drafted prose for human readers.
Watch out for: Do not rely on it to pass AI-detection checks; it is a style editor, not an evasion tool.

Assessed Oct 29, 2026 · adversarially reviewedSuggest a correction

Python · ≈66k stars · MIT