🐙
Freewritingclaude codeai detectionskillsopen source

Humanizer is a Claude Code skill that post-processes AI-generated text to remove the telltale patterns — passive constructions, robotic transitions, filler phrases — that make LLM output feel synthetic. Drop it into any agent workflow and it rewrites responses to match natural human writing cadence. With over 25,000 GitHub stars it has become a go-to tool for anyone who ships AI-generated copy to end users.

What is Humanizer?

Humanizer is a Claude Code skill that rewrites AI-generated text to remove the patterns that give it away. Models have habits, and after enough exposure they become impossible to unsee: the same hedging transitions, the same three-item lists, filler that restates the previous sentence, a uniform rhythm with no short sentences in it. Humanizer is a set of instructions the agent loads that tells it to strip those habits out of a draft and rewrite in a more natural cadence. It works as a post-processing pass, so you generate copy the way you normally would and then run it through, rather than adopting a different writing tool. Because a skill is a markdown file, you can read the guidance it applies and change it if the rules do not match your own voice. It installs into any agent workflow that loads Claude Code skills.

Who is Humanizer for?

The case for it is narrow and practical. You are generating a volume of prose that real people will read, marketing pages, documentation, release notes, outreach, and you do not have time to edit every draft line by line but you also cannot ship text that sounds like a model wrote it. A rewriting pass gets a draft closer to publishable before a human looks at it. It is worth being clear about what it is not: a rewrite changes surface style, it does not verify anything. If the draft contains a claim that is wrong, it will still be wrong afterward, only smoother. Keep the human review step for accuracy even when the prose reads well.

How does Humanizer compare?

Taste Skill and Impeccable are the nearest neighbors in intent, in that all three try to raise the quality of what an agent produces by giving it explicit standards rather than hoping the model has them. The difference is the medium. Impeccable governs interface design, with rules about layout, type and color that catch the visual defaults models fall into. Taste Skill addresses judgment more broadly. Humanizer works on prose alone. Marketing Skills overlaps with it from the other direction, since it packages copywriting workflows for agents, which is about producing the draft rather than cleaning it afterward. They stack fine: generate with one, pass through the other.

Compare with: Impeccable, Taste Skill, Marketing Skills

Humanizer pricing

The catalog records this as free. It is an open-source skill in a public repository, so there is nothing to buy and no account to create. Cost shows up only as model usage, since running it means a second pass over text you already generated, and rewriting long documents doubles the tokens spent on them. If you are processing a large volume of copy, that is the number worth watching. The skill file itself is small and adds little to your context.

Notes

No notes have been added for this tool yet.

Humanizer FAQ

Will this make text pass AI detectors?
Treat that as unproven. Detection tools are inconsistent, they change often, and no rewriting pass can promise a particular result against them. The reasonable expectation is prose that reads better to a human reader. If a specific detector outcome matters to you, test it yourself rather than relying on any claim.
How do I install a Claude Code skill?
Skills are markdown files placed in the directory your agent reads skills from, either inside a project or in your user configuration. Once the file is present, the agent loads it when the task matches. The repository documents the exact path and any supporting files it expects.
Does it change the meaning of my text?
A rewrite always carries some risk of drift, since removing filler and reshaping sentences touches the wording throughout. For anything with legal, medical or factual weight, read the output against the original. The skill improves style, it does not check whether the claims in the draft are correct.
Browse all Open Source