Stringers
adhiprib/humanizer-indonesia · adhiprib

An Indonesian lecturer with only 411 followers posted three Threads, and his 'AI writing de-stiffening' Skill got nearly 300 stars in 20 days

What it is A Claude Skill that makes AI-written Indonesian papers, textbooks and official documents read naturally, not stiff or like a translation. It also has ChatGPT and Gemini versions, free and open source.

Why it took off The author is a lecturer at an engineering college in Jakarta. He posted three Threads posts (peaking at 2.5K likes and 1.2K shares), which spread through university group chats. He positioned it against the grain, openly writing that it 'can't pass GPTZero', so teachers and students feel safe using and sharing it.

BWhat you could do Pick another major language you can write in yourself (Vietnamese, Thai, Brazilian Portuguese, Spanish, etc.) and focus on academic or official-document writing. Rewrite the rules for the local AI style, ship a ChatGPT / Gemini version and install video on day one, then post under your real name on the main local social platform with each release.

B grade: the playbook is clear and can be built over a weekend, and nobody has pushed the same-category library in each language yet. But the ceiling is a few hundred stars, and you have to post in that language under a credible real name, which most people can't do.Whispers in small circles
Data (updated daily) · Repo created 2026-10-24 · 345 stars · 7 days +48
GitHub · stars
010020030010-2510-3111-0711-1411-20345
Data 2026-10-25 → 2026-11-20 (27 days, 27 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

adhiprib/humanizer-indonesia is a Claude Skill that adapts the English hit blader/humanizer into an academic-writing version for Indonesian. It wasn't the first Indonesian version. At least 5 came before it, and none of them took off. This one did, and the reason isn't the code. It's who posted it: a lecturer at a polytechnic in Jakarta posted three times on Threads under his real name, stating plainly that "this won't pass AI detection." Below: what it is, where the stars came from, why this one, and whether this playbook still works for you.

A mechanical engineering lecturer's Threads posts

On Oct 24, 2026, Threads account @genshiryoku08 posted. The bio reads "pizza and ramen lover, coffee drinker, nuclear enthusiast," with 411 followers. His usual posts were everyday things like grading and peer review, e.g. "Reviewing a paper, got to this paragraph... sure enough, the PRISMA methods in the review were written by AI and nobody deleted it. I'd recommend rejecting." This one got 85 likes, his best post in a while.

That day he wrote:

"I made a skill for Claude and/or Antigravity (tried both) called humanizer-indonesia. The goal is to make AI-written text read more smoothly and naturally, not to get past AI detectors like GPTZero. Anyone who wants to try it can download it from GitHub."

The post got 707 likes, 135 reposts, and 258 shares (Threads page, 11-12 views). The author is Pribadi Mumpuni Adhi, who holds a doctorate in engineering from Tokyo Institute of Technology and is a mechanical engineering lecturer at Politeknik Negeri Jakarta (PNJ). You can verify this on the school website and Google Scholar. Before this, his GitHub account had no repositories at all.

What it is

A Claude Skill that rewrites AI-generated Indonesian text so it doesn't read stiff or like a translation. It's aimed mainly at papers, abstracts, project proposals, textbooks, lecture notes, and official documents.

  • 1000 lines of Indonesian rules: over a dozen categories of AI tells, including EYD V spelling conventions, KBBI dictionary usage, avoiding *saya* (I) in academic register, English-calque phrasing, stacked nominalizations, too many connectives, overly regular sentence rhythm, and "triplet" lists. The framework keeps the original English version's three-pass structure (draft, review, final), with all the rules rewritten.
  • Style presets: the author built a "my style" sample from his own articles written over 2008 to 2025 years, and the README encourages you to swap in your own. It also ships a metrics script using only the Python standard library, with three tiers: academic, textbook, and semi-formal.
  • Can review other people's drafts: you can use it to check text revised by Gemini or ChatGPT.
  • Works without Claude: a portable ChatGPT Project and Gemini Gem version came out the next day (10-25), and the author recorded two install tutorials on his own YouTube channel (425 and 325,11-12 views).

The first section of the README is titled "Important: this is not a tool for evading AI detection." It says it straight: text edited with it "will still be detected as AI-written." Automatic rewrites score only 1% to 2% human on GPTZero, and the only thing that clearly moves the score is the author editing by hand. The last line: "Use it honestly. If your school, journal, or supervisor requires disclosure of AI use, disclose it."

How the stars came in, day by day

TimelineStars (same-day snapshot)What happened
10-240Repo created, first Threads post (707 likes)
10-2554Second post: ChatGPT / Gemini portable version released, with YouTube tutorial (408 likes, 179 shares). The post says "already 81 stars and 13 forks"
10-27109This wave ended
10-28 to 11-07113 → 128Plateau, gaining 1 to 3 stars a day
11-07—v1.7 released, third post (2.5K likes, 579 reposts, 1.2K shares)
11-08192(+64)12 new forks that day
11-09 to 11-12203 → 2214 to 11 stars a day
11-13297(+76)No matching new post found

(GitHub daily snapshot; on the GitHub page 11-12 shows 299 stars and 39 forks.)

Every star surge came within 24 hours of a post, and stars barely moved when he wasn't posting. Searches on X, HN, Reddit, and TikTok turn up no one mentioning it, and no media coverage. The +76 on 11-13 has no public source. It may be the long tail of the v1.7 post, or it may have been forwarded in places search can't reach, like WhatsApp groups or Facebook (inference).

One detail shows where the posts reached: on all three, shares outnumber reposts, and the v1.7 post had 1.2K shares against 579 reposts. Lots of shares means people passed it privately into group chats, which fits circulation in university staff and student groups. In the comments someone directly tagged @Meta AI: "explain in simple words how to install this into Claude, ChatGPT and Gemini," and Meta AI's reply walked through downloading the zip and creating a Gem step by step. That reply got 2,400 views. The people looking at this are mostly teachers and students who don't know how to use GitHub.

Why this one, when it wasn't the first Indonesian version

GitHub already had Indonesian humanizers: onesoira/humanizer-bahasa-indonesia was created in March (17 stars), williamkho77/claude-humanizer-id in May (5 stars), plus three or four with 0 stars. The closest was nothinx/naskah-id, created 3 days earlier, also aimed at Indonesian academic writing, with fairly complete content, now at 106 stars. None of them got pushed.

The difference from these libraries lies less in how good the rules are and more in three points:

One, someone posted it, and the poster is credible. A lecturer releases a tool that "makes AI text read more human" himself. Students can use it without guilt, and colleagues can share it without embarrassment. The author had only 411 followers, which shows you don't need a big account first. It spread because people shared the content layer by layer.

Two, the positioning is the opposite. Most humanizer sellers on Indonesian TikTok shout "bypass Turnitin, guaranteed 100% human score," and charge per word or per subscription. He says the opposite: "won't pass GPTZero, disclose if you need to," and backs it with 14 measurements. In a university setting, that line makes it easier for lecturers and staff to share.

Three, ordinary people can install it. A ChatGPT and Gemini version arrived the next day, with a zip and video tutorials. GitHub was just the download page. The real users are on the Gemini and ChatGPT web apps. Around that time many Indonesian universities were running Gemini Academy training (news from late October), so "using Skills inside Gemini" was fresh for teachers.

"Localized humanizers" is already crowded, but the crowd is in repos, not in repos that are being pushed

The English original, blader/humanizer, is now at 68.5k stars. Localized versions followed in other languages, and several took off: Chinese op7418/Humanizer-zh (22.7k stars), Korean im-not-ai (6,500), Japanese yomiyasu (3,200), Traditional Chinese speak-human-tw (1,200), Russian (559), and Turkish (308,11-05, created only recently).

The remaining major languages aren't empty either. Nobody has pushed them yet. Below is the largest one 11-12 found on GitHub for each language:

LanguageLargest equivalentStars
Germanmarmbiz/humanizer-de205
Brazilian Portuguesemackswendhell/humanizer-pt-br67
Vietnameselonghang2004/vietnamese-humanizer62 (another viet-nhu-nguoi 11-07 was built, 46)
Spanishpeopleworks/SignsofAI31
Frenchferr079/humanizer-fr23
Thaicitelogics-th/thai-humanizer22
Arabic, Malay, Hindione or two each0 to 10

Same as Indonesian before October: the repos exist, the only missing piece is someone who posts under a real name on local platforms. The academic angle is also taken in English and Chinese (English academic-humanizer 4,676 stars, Chinese humanizer-zh-academic 371 stars).

If it were you

B: The playbook is clear and cheap, and you can build it over a weekend. But the ceiling is probably a few hundred stars, and it requires you to be able to use that language yourself and post under a credible identity. Most people can't do that.

Good fit for: developers whose native or working language is Vietnamese, Thai, Brazilian Portuguese, Spanish, Arabic, or Malay, ideally already in circles under writing pressure like universities, law firms, or hospitals, and who have a real-name account on the main local social platform (Threads in Indonesia, probably Facebook in Vietnam, Instagram and Threads in Brazil). People who want a project with a few hundred stars for their resume.

Not a good fit for: people who don't speak the language and just want to machine-translate a rule set and ship it (every language already has several of these, all in single-digit stars); people aiming for thousands of stars (outside Chinese, Korean, and Japanese, this category stalls at a few hundred); people wanting to do another Chinese or Indonesian version (someone has already made those take off).

How to do it:

  1. Swap the language, but don't just swap the language: rewrite the rules around that language's own AI tells. The Indonesian version has entries like "avoid *saya* in academic register" and "English-calque phrasing" that only a native writer would know to include. Then pick a specific scenario, such as academic writing, textbooks, or official documents, rather than building a general-purpose version.
  2. Ship a ChatGPT and Gemini version on day one: the audience is teachers and students, not developers. Record a phone-screen install video.
  3. Copy its honest positioning: state clearly at the top of the README that it's "not for beating detection," and include detection scores you measured yourself. In a market full of "guaranteed to pass" ads, that sentence is your biggest differentiator.
  4. Leave a style preset for people to swap: it has 39 forks and a fork rate of 13%, mostly because people swap in their own style. That makes people willing to bookmark it.

How to get users: post under your real name, you don't need followers first. Post with every release, and treat "what I updated" as the content. Write for teachers and students, not programmers, and don't mention stars. After posting, go through the comments one by one and show people how to install it.

When the window closes: new repos have already appeared for Vietnamese 11-07, Thai 10-24, and 10-27, so someone will eventually run this playbook in each language. My estimate is that before Q1 2027, whoever posts first will own the position in the major languages.

Signs you've got it wrong: ① your post goes viral (thousands of likes) but the repo still hasn't reached 50 stars a week later, which suggests local users paste prompts into Gemini and never click star on GitHub, so the star goal doesn't apply there; ② local education authorities or schools start publicly opposing AI humanizers, and lecturers stop sharing; ③ the reverse: someone does thousands of stars in Vietnamese or Brazilian Portuguese with the same playbook, which means the ceiling is higher than estimated here and you should move faster.

GitHub repo ↗

Research notes: competitor table, acquisition evidence, full research report
Competitors
NameWhereLaunchedScaleHow it differs from this one
blader/humanizerOpen source2026-01-1868.5k★The English original, with rules based on Wikipedia's 'Signs of AI writing.' This library keeps its framework, rewrites all rules in Indonesian, and adds style presets, a metrics script, and GPTZero measurements.
nothinx/naskah-idOpen source2026-10-21106★Also aimed at Indonesian academic writing (EYD / KBBI), focused on normative writing and proofreading. No social distribution and no ChatGPT / Gemini portable version.
FRCHN-GSTN/MenulisCep4tOpen source2026-10-2334★Bundles naskah-id and blader into an academic document engine (Typst / Pandoc / MCP). Engineered, with a high barrier to entry.
onesoira/humanizer-bahasa-indonesiaOpen source2026-03-1117★An Indonesian humanizer released 7 months earlier, supporting only Claude Code / Codex CLI, with no deep academic-register rules and no distribution.
williamkho77/claude-humanizer-id and others (ardha27, ilhamfzri, saifulveri)Open source2026-05~100–5★Direct ports of blader or shallow social-media casual style. Thin content, nobody pushing them.
op7418/Humanizer-zh, epoko77-ai/im-not-ai, nanaism/yomiyasuOpen source2026-01~0922.7k / 6.5k / 3.2k★Localized versions for Chinese, Korean, and Japanese that prove the playbook works. Different languages, so no direct competition.
QuillBot AI Humanizer / Undetectable.ai / Lynote Indonesian humanizer / jurnalhub.idWeb2023–2026quillbot.com Tranco 3,680;undetectable.ai 58,165;lynote.ai 333,721Commercial, pitched as getting past Turnitin / GPTZero; this repo is free and open source, and openly says it won't pass detection, it only aims for readability
ChatGPT / Grammarly AI Humanizer, or just ask Claude / Gemini to rewrite itBig-company features2025–2026HugeGeneral rewriting, with no Indonesian academic style rules, style presets, or audit process
Where users come from
  • Verified Threads author's launch post: 10-24 @genshiryoku08 posted about humanizer-indonesia, 707 likes / 47 replies / 135 reposts / 258 shares; 10-25 to 10-27 stars went from 0 to 109
  • Verified Threads v1.7 release post: about 11-07 posts, 2.5K likes / 579 reposts / 1.2K shares; 11-07 UTC new forks 12, 11-08 stars +64. The author's usual posts only get 1–85 likes
  • Verified Threads quote reposts / Meta AI replies: @karyawanjuga, @ferrrrrrr_2, @_praynadeak, @appsmurahbgt_ quote-reposted it; Meta AI answered the install steps in the comments
  • Verified Author's YouTube tutorial: Prib's Lab install tutorials for ChatGPT / Gemini, 425 / 325 views
  • Inferred Private group-chat forwarding (WhatsApp / university groups): the shares on the two Threads posts (258, 1.2K) are far higher than the reposts; 11-13 +76 with no matching public post found
  • Verified Passive listing on a skill directory: listed at qai.io/skills/adhiprib/humanizer-indonesia

adhiprib/humanizer-indonesia: why it took off

Channel: GitHub; data as of 2026-11-13 (our GitHub repo_day). Live GitHub check: 2026-11-12:299 stars, 39 forks.

Bottom line

An Indonesian state polytechnic lecturer took the verified English blader/humanizer, adapted it for Indonesian academic writing, and posted two threads on his own Threads account (10-24 first post 707 likes, 11-07 post v1.7 with 2.5K likes / 579 reposts / 1.2K shares). Both star jumps came within 24 hours of posting; GitHub was just the landing page. It took off for three reasons: Threads has heavy traffic in Indonesia, the users are Indonesian university students and staff, and the author is credible and honest in writing that it "gets past GPTZero". Calling it "the best in its category" isn't quite accurate. The more precise claim is that among Indonesian-language humanizers, it's the only one that actually got distributed.

What it is

  • A Claude Skill (SKILL.md, 1089 lines), plus references/id-ai-tells.md (a catalog of Indonesian AI writing tells), references/voice-prib-id.md (a "style preset" built from the author's own 2008–2025 years of writing), and scripts/metrik.py (Indonesian style metrics using only the Python standard library, with three presets: akademik / ajar / semiformal).
  • From v1.6 on it ships portable/, so ChatGPT Projects / Custom GPTs and Gemini Gems work too. There's also a written install guide for Google Antigravity.
  • Use cases: theses, abstracts, conference papers, grant proposals, textbooks, lecture notes, reports, emails. Legal documents, direct quotes, and validated research instruments are out of scope.
  • The positioning is blunt: the intro literally says "Membuat tulisan AI lebih luwes dan enak dibaca, bukan untuk mengelabui deteksi AI" (to make AI text smoother and easier to read, not to fool AI detection). The README publishes 14 GPTZero measurements: automated rewriting left human scores at only 1–2%, and the only thing that actually moved the score was the author's own hand edits.
  • Author Pribadi Mumpuni Adhi (Dr.Eng., PhD from Tokyo Tech), lecturer in mechanical engineering at Politeknik Negeri Jakarta (PNJ). He has a Google Scholar profile and an archived PNJ faculty page. His GitHub account has only this one repo, and his Threads account is @genshiryoku08 ("Nuclear Enthusiast"). Sources: mesin.pnj.ac.id/readmore/5e2016a01c6cc70c3069bca0, scholar.google.com/citations?user=gFFCNL8AAAAJ

Growth data (our GitHub repo_day + GitHub fork timestamps)

TimelinestarEvent
10-24 05:20 UTC0Repo created, v1.5 released; first Threads post the same day
10-25547 forks on day one
10-27109The wave from the first post mostly ended; forks at 18
10-28 → 11-07113 → 128Plateau, 1–3 per day
11-06 23:51 UTC—v1.7 released, Threads v1.7 post
11-08192(+64)11-07 (UTC) new forks: 12
11-09 → 11-12203 → 221Back down to 4–11 per day
11-13297(+76)Third wave; 11-12 UTC new forks: 5, trigger unverified
  • Over the last 7 days it gained 170, versus only 7 in the first 11 days; single-day peak +76 (11-13). Forks/stars about 13%. It hasn't made trending yet, but a repo less than 60 days old couldn't have made it anyway, so that proves nothing.
  • Release attachment downloads: the two v1.7 zips were downloaded 88 and 78 times, and v1.6.1 66 times (GitHub API, 2026-11-12). Most people downloading the zip are probably Claude.ai / Gemini web users, which means a real group of people actually installed it.
  • The author's own YouTube channel, "Prib's Lab", has one ChatGPT install tutorial and one Gemini install tutorial, with 425 and 325 views (read via Jina, 2026-11-12).

Where users come from (evidence)

  1. Threads launch post (10-24): "Saya membuat skill untuk claude dan/atau antigravity … humanizer-indonesia. Tujuan skill ini untuk membuat tulisan AI lebih luwes dan enak dibaca, bukan untuk mengakali lolos uji AI seperti GPTZero." Data: 707 likes, 47 replies, 135 reposts, 258 shares. www.threads.com/@genshiryoku08/post/De3Y1V3k_TH (quoted by @karyawanjuga and @ferrrrrrr_2, visible on the page)
  2. Author replied under someone else's humanizer post: "Bang saya kembangin jadi humanizer Indonesia" (10-24,102 views, very small reach). www.threads.com/@genshiryoku08/post/De3i-xdE7U7
  3. Threads v1.7 post (about 11-07): "Humanizer-indonesia versi 1.7 sudah saya rilis di GitHub … terutama tulisan akademik dan bahan ajar … sejak versi 1.6 juga dapat dipakai di ChatGPT dan Gemini." Data: 2.5K likes, 47 replies, 579 reposts, 1.2K shares. www.threads.com/@genshiryoku08/post/Dfa1st1j6i3. The author's usual posts only get 1–85 likes (on his profile, the calculator app and the peer-review posts), so this one clearly broke out.
  4. Meta AI answered install questions under the author's Threads post on his behalf ("Izin jelasin simpel ya pak @genshiryoku08: 1. Claude … 2. ChatGPT & Gemini …"), which suggests the comments are full of non-technical users who can't install it. www.threads.com/@meta.ai/post/DfbNRHMjAtv
  5. The post was still being quote-reposted several days after it went up (@_praynadeak 5 days ago, @appsmurahbgt_ 1 days ago). The +76 at 11-13 may be this long tail, or it may come from places we can't search, like Facebook or WhatsApp groups. Unverified.
  6. Not found: X (x-radar searched "humanizer indonesia" over the last 30 days and got 0 results), TikTok (TikHub search for "humanizer indonesia" only returned videos about commercial humanizers), HN, Reddit, and press coverage. The QAI skill directory (qai.io/skills/adhiprib/humanizer-indonesia) lists it automatically, which counts as passive exposure.

Competitors

Open source: same-category Indonesian projects (all small)

NameLaunchedScaleHow this repo differs
nothinx/naskah-id2026-10-21106★Also targets Indonesian academic writing (EYD V / KBBI / SINTA / thesis writing), 3 days older than this repo, with a more "standard writing and proofreading" positioning. No social distribution by its author found.
FRCHN-GSTN/MenulisCep4t2026-10-2334★Combines naskah-id and blader, plus Typst / Pandoc campus templates and MCP. Engineering-heavy, high barrier to entry.
onesoira (formerly ariosenodev) / humanizer-bahasa-indonesia2026-03-1117★7 months earlier, with 24 general patterns + 9 Indonesian patterns, supporting only Claude Code / Codex CLI. Listed on skills.rest.
williamkho77/claude-humanizer-id2026-05-065★Direct port of blader
ardha27/humanizer-id, ilhamfzri/humanizer-indonesia, saifulveri/humanizer2026-08~100★Lean toward casual social-media language or are direct forks of blader

This repo is not the first in the Indonesian-language space. At least 5 came before it, and none of them took off.

Open source: same-category projects in other languages (showing that "localized humanizer skill" as a play has been validated repeatedly) blader/humanizer 68.5k★ (2026-01); op7418/Humanizer-zh 22.7k★ (2026-01); epoko77-ai/im-not-ai (Korean) 6.5k★; nanaism/yomiyasu (Japanese) 3.2k★; Raymondhou0917/speak-human-tw 1.2k★; ilyautov/humanizer-ru 559★; yucezerey/turkish-anti-ai-skill 308★ (11-05 repo created). Data from our GitHub repo and the GitHub API, 2026-11-12.

Web / commercial: QuillBot AI Humanizer (Tranco 3,680), Undetectable.ai (58,165), Lynote's ai-humanizer-indonesia page (lynote.ai 333,721), Humbot (405,209), Tenorshare, SupWriter, jurnalhub.id (an Indonesian AI portal aimed at students and lecturers; a TikTok post hit 11k views). All of these sell "bypass Turnitin / GPTZero," charging a fee or per word. Tranco ranks come from our traffic library 2026-11-12.

Big-tech features: ChatGPT and Grammarly each have their own AI Humanizer pages, and Claude and Gemini can make text "sound more natural" if you just ask, but none has rules tailored to academic Indonesian style.

Why this one took off

  1. Distribution: the author's own Threads posts took off. Both star jumps followed two posts closely. The first post got 258 shares, and the v1.7 post got 1.2K shares. Lots of "shares" suggests it was forwarded privately into group chats, which matches the card's guess that it spread through university group chats. No comparable library had this kind of distribution.
  2. A credible persona. The poster is a real university lecturer and peer reviewer. His usual posts cover grading assignments and flagging AI use in PRISMA reviews. A lecturer personally releasing a tool that "makes AI text read more like human writing" carries endorsement and feels "safe" to students and peers.
  3. Positioning against the grain works well in the Indonesian context. Indonesian TikTok humanizers almost all sell "beat the detector, guaranteed 100% human score." He openly writes "can't pass GPTZero, disclose it if you must," and attaches 14 measurements. That makes it safe for lecturers and institutions to share, and students find it more trustworthy.
  4. The barrier to entry dropped to ordinary-user level. From v1.6 on, it ships portable versions for ChatGPT and Gemini, with zip downloads, YouTube tutorials, and Meta AI answering comments on his behalf. The second wave was bigger than the first, partly because the audience widened from Claude users to Gemini and ChatGPT users. It also coincided with Indonesian universities running a wave of Gemini Academy training (UPNVJ, Unhas, SCU and others, news from late October). Eudeka 11-03 also published a tutorial on how teachers and lecturers can use Skills in Gemini.
  5. The product is substantial. 1000 lines of rules cover more than a dozen categories of AI tells: EYD V / KBBI spelling conventions, avoiding *saya* in academic register, English calque (kalke) phrasing, "triple parallelism," and more. It can also audit text revised by Gemini or ChatGPT, and ships presets that can be swapped for your own writing style, plus metric scripts. People who see something this complete are willing to star and bookmark it, and to fork it to swap in their own style presets (the README explicitly encourages this, which explains the high fork rate).

Differentiation

  • Compared with the English original and localized ports in other languages: the rules are fully rewritten for Indonesian, with a specialized focus on academic and textbook register. It also has style presets calibrated on the author's real corpus, zero-dependency metric scripts, an "audit other models' revisions" step, and publicly verifiable GPTZero measurement records.
  • Compared with open-source Indonesian alternatives: the content is roughly on par with naskah-id. The real differences are twofold. First, portable multi-platform versions plus video tutorials, which cover most people who don't use Claude Code. Second, the author personally posting on Threads.
  • Compared with commercial humanizers: it's free, open source, and openly admits it can't pass detection.
  • Were these differentiators the reason it took off? Only half. Content quality is what sustained "worth sharing" and "worth forking." But what made it take off rather than naskah-id or onesoira was distribution (the Threads posts), persona (a real lecturer), and portable versions together.

Visibility

It's circulating in small circles. Two hit posts on Indonesian Threads (up to 2.5K likes). No media coverage, no discussion on X, HN, or Reddit, and no revenue to speak of (purely open source, no monetization signals). Viewed from the global open-source scene, almost nobody knows about it, and people outside Indonesian-speaking circles are unlikely to notice.

Copycats

No clear copycats found. naskah-id (10-21) and MenulisCep4t (10-23) both came earlier, and I found no Indonesian humanizer library created after it. The 39 forks are mainly individuals swapping in their own style presets.

What others can borrow

  • The approach of "a verified English skill × a major language × a specific audience (universities, academia)" has worked in Chinese, Korean, Japanese, Traditional Chinese, Russian, and Turkish. Indonesian has a large population, yet no one broke out with this until 2026-10. Academic editions in Vietnamese, Thai, Portuguese (Brazil), and Spanish are good places to look.
  • The method matters more than the tool. The author posts under his real identity on the country's dominant social platform (Threads in Indonesia), and each version release gets its own post, treating "version updates" as recurring content.
  • Portable versions (instruction packs for ChatGPT Projects / Gemini Gems) plus YouTube tutorials expand the skill's audience from developers to ordinary users.
  • Positioning against the grain in the gray market of "bypassing detection," stating plainly "this is not for passing detection" and publishing the measurements openly, earns the trust that lets institutions and lecturers share it.

Still unclear

  • Where did the +76 in 11-13 come from? I didn't find new posts. It may be the long tail of the v1.7 post, or private forwarding through Facebook, WhatsApp groups, and similar channels.
  • GitHub's stargazers endpoint returned 401/403 to us, so we can't match stars to posts by the hour. For now the only option is inferring from fork timestamps and daily snapshots.
  • Real usage: about 230 release downloads. Git clones and Gem pastes are unknown.
  • Threads post view counts aren't shown on the page, only likes, reposts, and shares.
  • Note: the "today" in why.md is 2026-11-07, while the data in the data card and this report is as of 2026-11-13.

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.