STE80 Humanizer review (2026): an honest look at the free, open-source agent skill that rewrites AI text toward ASD-STE100 clarity — and who should skip it.
STE80 Humanizer is one of the most honest tools in the "humanizer" category — because it refuses the category's usual job. It does not try to beat detectors; it rewrites AI text to read clearly, using ASD-STE100-inspired structure rules, and it does that narrow job well. It is free, MIT-licensed, and lives inside your coding agent. Install it if you write explanatory prose and want it tighter; it will not create or publish content, so it is no substitute for a content engine.
STE80 Humanizer arrived in late 2026 riding a specific idea: Andrej Karpathy's October 2, 2026 suggestion that language models write "about 80% of the way to ASD-STE100 Simplified Technical English" — the controlled English built for aircraft maintenance manuals — because the constraints produce more readable text. The skill turns that one-liner into a repeatable tool: a two-pass rewrite that strips AI tells and applies simplified-English structure rules, packaged as an open-source agent skill with a small Python linter.
This review judges STE80 on its own stated terms, which matters here more than usual, because the word "humanizer" sets a false expectation. Nearly every product with that label exists to evade AI detectors. STE80 explicitly says detector evasion is not a goal; the aim is readability. So the questions are different: does it actually make text clearer, how faithful is it to the ASD-STE100 spirit, is it pleasant to run, and is it worth adding to your workflow. The short answer is that it is a genuinely good, genuinely narrow tool, and it is unusually honest about being narrow.
I am not going to inflate its weaknesses to flatter a content engine. STE80 and a publishing platform are not in the same lane, and STE80 never claims to be. What it claims — "I will make this block of English read more cleanly without changing the meaning" — it largely delivers, for free, under an MIT license. The caveats are about scope, language, and the difference between "STE style" and real STE compliance, not about whether the tool works.
STE80 Humanizer is an open-source "agent skill" published on GitHub by the account Dodaios, under the MIT license. Rather than a website, it installs into coding agents — Claude Code, Codex, Cursor, and Claude.ai/Desktop — and also ships a dependency-free Python linter (scripts/ste_check.py). You give it AI-generated or rough text and it rewrites it to read more like a person wrote it, without changing the meaning. Mechanically it runs two passes. Pass 1 removes 26 "AI tells" sourced 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, and so on. Three modes set the intensity: Humanize (default), Light (Pass 1 plus only the worst Pass 2 offenders), and Strict STE (all of Pass 2). It can match a voice from a sample you supply and runs a fact-fidelity focus that preserves facts and flags contradictions. Importantly, it paraphrases the ASD-STE100 rules and ships no dictionary or specification, is not affiliated with ASD, and states that "STE style" is not formal "STE compliance," which requires the official dictionary, trained writers, and review.
STE80 is for people who already work inside a coding agent and produce explanatory English prose — developers writing READMEs, docs, and changelogs; technical and how-to writers; marketers drafting clear blog explainers or product copy; and non-native English writers who want tighter, less ambiguous sentences. It fits best when the text is expository, English, and you want a fast, free, local pass that makes it cleaner. It is a poor fit for anyone who needs to generate content (images, video, carousels), enforce a brand voice across a batch, publish to platforms, or evade an AI detector — none of which it does or claims to do.
| Dimension | Score | Why |
|---|---|---|
| Readability rewrite quality | 4.2 / 5 | The core job — clearer, less ambiguous sentences without altering meaning — is handled well on expository text. |
| AI-tell removal (Pass 1) | 4.3 / 5 | Strips the recognizable openers, padding, and negative-parallelism tics drawn from well-known catalogs. |
| ASD-STE100 fidelity | 3.5 / 5 | Paraphrased structure rules in the STE spirit; honestly not "STE compliant" (no dictionary, no certified review). |
| Workflow & install | 3.8 / 5 | One-line install into agents you may already use, but it requires an agent — there is no standalone GUI or web app. |
| Transparency & honesty | 4.8 / 5 | Rare candor: disclaims detector evasion, disclaims compliance, credits its sources, and notes its linter is a heuristic. |
| Language support | 2.9 / 5 | The STE structure rules are English-only; other languages get the AI-tell and generic structure passes at best. |
| Breadth beyond rewriting | 1.5 / 5 | Does exactly one thing. No generation, no brand governance across formats, no scheduling or publishing. |
| Value for money | 4.9 / 5 | Free and MIT-licensed with no dictionary to license — nothing to pay, easy to try, easy to remove. |
There is almost nothing to analyze, and that is the point. STE80 Humanizer is free and MIT-licensed. There is no paid tier, no credit meter, and no dictionary to license, because it paraphrases the ASD-STE100 rules rather than bundling the (separately free) specification. You install it in seconds and remove it just as fast. For a tool whose whole promise is "make this text clearer," free is exactly the right price and removes any reason not to try it.
The honest caveat is the hidden cost of the category it sits in: it only earns its keep if you already live in a coding agent and already produce the kind of expository English it is tuned for. If you do, the value is effectively infinite relative to zero dollars. If you do not — if you are a creator who needs to produce and ship on-brand video, images, and posts — the price being free does not change the fact that it does not do your job.
Compared against paid humanizers (Ryne AI, UltraHumanizer, and the like) the comparison is almost unfair in STE80's favor on price, but they are not substitutes: those tools charge for detector evasion, which STE80 refuses to attempt. And compared against a content engine, it is apples to oranges — STE80 at $0 buys a clarity pass; a generation-and-publishing platform buys the ability to create and distribute content. Neither can be bought with the other.
| Use case | Fit | Why |
|---|---|---|
| Tightening READMEs, docs, and changelogs in your editor | Strong | Expository English inside a coding agent is its ideal input; the structure rules shine here. |
| Cleaning an AI-drafted blog explainer before it ships | Strong | Pass 1 removes the AI tells and Pass 2 shortens and clarifies — exactly the readability edit the draft needs. |
| Clearer writing for a non-native English author | Strong | Controlled-English habits — one idea per sentence, approved-style vocabulary — reduce ambiguity for ESL writers. |
| Passing an AI detector like Turnitin or GPTZero | Weak | An explicit non-goal; the project disclaims detector evasion and its linter is not a detector. |
| Certified ASD-STE100 for regulated aerospace/technical docs | Weak | "STE style" is not compliance — that needs the official dictionary, trained writers, and formal review. |
| Punching up fiction, speeches, or brand voice copy | OK | Only the AI-tell pass applies by design; the STE structure rules would flatten creative or persuasive writing. |
| Producing and publishing content across platforms | Weak | It rewrites text only — no generation of media, no brand governance across formats, no scheduler. |
The honest comparison is that STE80 and Kompozy barely overlap, and pretending otherwise would fail the sniff test. STE80 rewrites one block of existing English toward clarity, for free, inside your editor. Kompozy is a generation-and-publishing engine: it writes net-new copy, renders video, images, carousels, blogs, and newsletters, and schedules them across the eight social platforms plus blog and email.
Where they do touch is the principle, not the feature. STE80's best idea — specify a clean, controlled style and apply it — is something Kompozy does at generation time rather than as a cleanup pass: the Persona Brief fixes your voice and a banned-word filter strips the same AI tells STE80's Pass 1 targets, so clarity is a property of every draft. If your only need is to make a paragraph read better, STE80 is the right, cheaper tool and you should use it. If your need is to produce and distribute on-brand content at volume, Kompozy is the category you actually want — and the two can coexist, with STE80 polishing a one-off block and Kompozy governing the rest.
Yes. It is open-source under the MIT license, with no paid tier and no subscription. It installs into coding agents (Claude Code, Codex, Cursor, Claude.ai/Desktop) and ships a dependency-free Python linter. There is no dictionary to license because it paraphrases the ASD-STE100 rules rather than bundling the specification.
No — and it says so directly. Unlike most tools labeled "humanizer," STE80 states that getting past AI detectors is not a goal; the aim is readability. Its checker is described as a heuristic, not an AI detector and not a compliance checker. If detector evasion is your goal, STE80 is the wrong tool.
No, and the project is clear about it. It produces "STE style," not formal "STE compliance." Real compliance requires the official ASD-STE100 dictionary, trained writers, and formal review. STE80 paraphrases the structure rules in the spirit of the standard and ships no dictionary.
People who work inside a coding agent and write explanatory English — developers, technical and how-to writers, marketers drafting clear blog copy, and non-native English writers. It is a poor fit for anyone who needs to generate media, enforce a brand voice across a batch, publish to platforms, or evade a detector.
Framing and scope. Ryne AI and similar tools charge to make AI text score "human" on detectors; STE80 refuses detector evasion and optimizes readability instead, for free. Both are text-only rewriters — neither creates new content or publishes. For those jobs you want a content engine like Kompozy.
No. It rewrites a block of text you already have. It generates no images, video, carousels, blogs, or newsletters, enforces no persistent brand voice across formats, and has no scheduler or publishing. A content engine such as Kompozy covers generation and multi-platform publishing; STE80 covers the narrow clarity edit.
It comes from Andrej Karpathy's October 2, 2026 suggestion to prompt models to write "about 80% of the way to ASD-STE100 Simplified Technical English." The full standard is strict, so softening it to roughly 80% keeps the clarity gains with a more natural flow. The skill turns that idea into a repeatable two-pass rewrite.