Free, open-source agent skill that rewrites AI-generated text toward ASD-STE100 clarity — explicitly for readability, not detector evasion.
Last verified · 2026-11-10 · by Moe Ameen
STE80 Humanizer is an open-source "agent skill" published on GitHub by the account Dodaios under the MIT license. Instead of a website, it installs into coding agents — Claude Code, Codex, Cursor, and Claude.ai/Desktop — and also ships a small dependency-free Python linter (scripts/ste_check.py). You hand it AI-generated or rough English and it rewrites the text to read more like a person wrote it, without changing the meaning.
It implements an idea Andrej Karpathy posted on October 2, 2026: prompt a model to write "about 80% of the way to ASD-STE100 Simplified Technical English," the controlled English originally written for aircraft maintenance manuals, because its constraints produce cleaner, more readable prose. The full standard is strict, so the "80%" softening keeps the clarity with a more natural flow — and gives the skill its name. ASD-STE100 is maintained by the ASD Simplified Technical English Maintenance Group, pairs writing rules with a dictionary of roughly 900 approved words, and reached Issue 9 in January 2025.
Under the hood it runs two passes. Pass 1 removes 26 "AI tells" drawn from Wikipedia's "Signs of AI writing" and Siqi Chen's MIT-licensed blader/humanizer catalog. Pass 2 applies 15 ASD-STE100-inspired structure rules — one idea per sentence, short sentences, active voice. Three modes set the intensity (Humanize, Light, Strict STE), and it can match a voice from a sample and run a fact-fidelity check that flags contradictions with the source.
Two things are worth stating plainly. First, it is explicit that getting past AI detectors is not a goal — the aim is readability — and its linter is a heuristic, "not an AI detector and not a compliance checker." Second, it edits one block of English text: it paraphrases the ASD-STE100 rules (shipping no dictionary), produces "STE style" rather than certified compliance, and generates no images, video, or finished posts and publishes nothing.
STE80 is a drafting-stage tool. Picture your pipeline as three stages — write, polish, ship — and STE80 owns the middle one: it takes a rough or AI-drafted block of English and makes it read cleanly, right inside the agent you were already working in. What it does not have is a next stage. The polished paragraph sits in your editor; it still has to become a caption, a carousel, a script, a newsletter, and then get onto feeds. [Kompozy](/) is the ship stage.
The concrete workflow: draft and run STE80 on your core copy — a product explainer, a hook, a blog section — then paste that clean text into Kompozy as a source. From one clarified idea Kompozy generates the formats STE80 structurally can't: brand-exact [Carousel Posts](/glossary/hyperframes) and quote cards rendered through HyperFrames, a [Persona Short](/glossary/persona-shorts) where a recurring avatar narrates the point to camera, a Listicle Video, a full blog article, and an email newsletter. A [Persona Brief](/glossary/persona-brief) holds the voice across all of them and a banned-word filter keeps the same AI tells out of the net-new copy Kompozy writes — so the clarity you bought with STE80 in one paragraph becomes the standard for the whole batch. Then [Autopilot](/glossary/autopilot) schedules and publishes the set across the eight social platforms plus blog and email, with a per-post review gate where you approve each piece. STE80 makes the sentence clear; Kompozy turns that sentence into a week of on-brand, multi-platform content and puts it live. If you reached STE80 because you wanted cleaner AI copy, see also the honest [STE80 Humanizer review](/reviews/ste80-humanizer) and how it stacks up as a [Kompozy alternative](/alternatives/ste80-humanizer), plus the detector-evasion humanizers it deliberately isn't — [Ryne AI](/ai-tools/ryne-ai) and [UltraHumanizer](/ai-tools/ultrahumanizer).
STE80 Humanizer is a free, open-source (MIT) agent skill published by Dodaios (a GitHub organization) that rewrites AI-generated English to read more naturally without changing its meaning. It installs into coding agents — Claude Code, Codex, Cursor, and Claude.ai/Desktop — and runs a two-pass rewrite: one pass removes common "AI tells," the other applies ASD-STE100-inspired structure rules like one idea per sentence and active voice.
No. The project states explicitly that getting past AI detectors is not a goal — it optimizes for readability instead, and its linter is a heuristic rather than a detector. If your goal is detector evasion, STE80 is the wrong tool (and the detector-evasion category is fragile and contested).
It is free under the MIT license, with no paid tier. It is distributed as an agent skill, so you run it inside a supported agent (Claude Code, Codex, Cursor, or Claude.ai/Desktop); it also ships a dependency-free Python linter. There is no standalone web app or GUI.
No. It rewrites a block of text you already have. It generates no images, video, carousels, blogs, or newsletters, enforces no brand voice across formats, and has no scheduler or publishing. To turn clarified copy into multi-format content and publish it across platforms, pair it with a content engine like Kompozy.