Humanizer Indonesia skill
A skill that edits Indonesian text: removes machine phrasing and English calques while keeping the author's voice
- 299 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skill reads and edits the texts you give it and runs a local Python metrics script
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/humanizer-indonesiaDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill reads and edits the texts you give it and runs a local Python metrics script
This is third-party code. Review the repository files before installing.
What it does
Humanizer Indonesia is a Claude skill for editing Indonesian text. It looks for stacked nominalizations, redundant connectives, uniform sentence rhythm and calques from English, then proposes edits in the author's voice. It follows the eight steps in SKILL.md and can also audit a file another model, such as Gemini or ChatGPT, has already edited. The repository holds a catalog of machine-text tells in Indonesian, a scripts/metrik.py style-metrics script built on the Python standard library, and a sample voice preset. You can swap the preset for your own; without one the skill falls back to a neutral voice and says plainly that no voice calibration was done. A simplified version for ChatGPT and Gemini sits in the portable folder. The author stresses that the skill does not beat AI detectors: in GPTZero measurements edited text was still flagged as machine-written. The output is a draft for manual editing.
Who it is for. Authors, students and editors who write in Indonesian and rework drafts produced by a model or translated from English.
Good fit when
- An Indonesian draft reads like a translation and you need to cut calques and heavy constructions
- A text was already edited by Gemini or ChatGPT and you want an independent audit of the edit
- You want to keep your own voice, matched to earlier writing
- An article or abstract needs a livelier rhythm before submission
Not a fit when
- You need to pass an AI detector: the author states the skill does not work for that
- The text is in another language: the tells catalog and metrics target Indonesian
- You need a publish-ready text without manual review: the skill returns a draft
Example request
Use the humanizer-indonesia skill and edit this paragraph so it does not sound like a translationLimitations
The skill gives no guarantee against AI detectors. The bundled voice preset comes from the author's own texts, so replace it to match your style. The metrics script needs Python 3 and has no other dependencies. The skill is built for Indonesian; the metric thresholds and the tells catalog do not fit other languages. Always read the result yourself, since edits change tone and word order. If your institution or journal requires disclosing AI use, you have to do that yourself.
How to disable. Delete the humanizer-indonesia folder from your skills directory (~/.claude/skills/ for Claude Code). If it was installed in several places, remove every copy so the agent does not load two versions.
FAQ
Does the skill get past AI detectors?
No. The author says so at the top of the README and cites measurements: after editing, GPTZero still rates the text as machine-written. The skill improves readability; manual editing remains yours.
How do I install it in Claude Code?
Clone the repository with git clone into ~/.claude/skills/humanizer-indonesia. For Claude.ai, download the ZIP and upload it in Skills settings, with the humanizer-indonesia folder at the archive root.
Does it work with Antigravity or ChatGPT?
Antigravity reads the same SKILL.md format: place the folder in ~/.gemini/config/skills/ or the project's .agents/skills/. For ChatGPT and Gemini on the web there is a version in the portable folder with a short instruction and a package-building script.
Can I replace the voice preset?
Yes. references/voice-prib-id.md is the author's sample preset, built from his texts of 2008 to 2025. Put your own in its place, describing rhythm, vocabulary and favorite phrasing, and edits will follow your style.
What is metrik.py?
It is a style meter for Indonesian text, written on the Python standard library. The agent runs it in a terminal, so Python 3 must be installed. The author calibrated the thresholds against his own corpus.
Discussion
No questions or reviews yet.
Related
SkillsSkill and CLI auditor for editing academic Indonesian text: removes calques, stiffness and cliches while leaving facts intact.
The Chinese version of Humanizer: finds and removes signs of AI writing in Chinese text
A skill that finds AI writing tells, rewrites the text in a chosen voice, and optionally scores it from 0 to 100
Humanizer (brandonwise)
humanizer
A skill and Node.js CLI that score English text for AI-writing patterns and statistics and suggest fixes on a 0 to 100 scale
A skill that removes AI tone from work writing while keeping every fact, commitment and caveat at the same strength.
A skill for Claude Code, Codex and Cursor: strips signs of AI writing and checks the result against human-writing ranges.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Humanizer Indonesia (this card) | no | 299 | |
| Humanizer | no | 68,355 | edits English text, while this skill has a tells catalog for Indonesian |
| Humanizer-zh | no | 22,682 | the same for Chinese |
README in short
The README opens with a warning: the skill improves style and does not disguise machine authorship. It continues with a table of repository contents, install routes for Claude.ai, Claude Code, Antigravity and ChatGPT with Gemini, a section on replacing the voice preset and a version history, the latest being 1.7.
README badge
Original SKILL.md
--- name: humanizer-indonesia description: "Menyunting teks bahasa Indonesia agar tidak terbaca sebagai tulisan AI atau terjemahan, dan kembali ke gaya penulis. Juga untuk mengaudit berkas yang sudah disunting model lain. Trigger: /humanizer-id, humanize, rapikan bahasa, biar tidak kaku, lebih mengalir, cek gaya, audit hasil Gemini." --- # Humanizer Indonesia **Versi 1.7, 26 Oktober 2026.** Untuk memastikan versi mana yang aktif, tanya saja "humanizer-indonesia versi berapa" dan jawabannya diambil dari baris ini. Riwayat singkat. 1.0 rilis awal. 1.1 memperbaiki deteksi verba ringan dan menambah metrik kalimat panjang. 1.2 mengkalibrasi ulang seluruh ambang ke rentang korpus yang sebenarnya, setelah terbukti ambang 1.0 dan 1.1 menghukum tulisan manusia penulis sendiri, dan menetapkan bahwa suara penulis menang atas skor detektor. 1.3 menegaskan posisi skill sebagai penyiap draf untuk disunting manusia, melengkapi matriks bukti detektor tujuh pengukuran, dan mengoreksi rentang diagnostik kepadatan pemadat ke 73 sampai 139. 1.4 menambahkan Langkah 5b tentang kalke dan diksi sebagai langkah wajib, setelah satu bab diktat lulus sembilan dari sembilan metrik tetapi tetap terbaca sebagai terjemahan, dan menambahkan bagian kaidah yang sering terlewat setelah aturan tanda pisah terbukti keliru diterapkan sampai menghapus tanda hubung pada bentuk ulang. 1.5 menambahkan Langkah 5c tentang personifikasi, register kolokial, dan pembuka kalimat monoton, menambahkan Langkah 5d tentang fusi kalimat terfragmentasi, menambahkan bagian audit untuk berkas yang sudah disunting model lain, serta membetulkan ambang kepadatan pemadat pada Langkah 7 yang masih tertulis 45 padahal seharusnya 73. Seluruh tambahan itu lahir dari perbandingan langsung dua bab diktat versi Claude lawan versi Gemini pada 9 September 2026, ketika penulis menilai bahasa Indonesia versi Gemini lebih baik sedangkan versi Gemini ternyata menghapus rujukan silang dan angka terverifikasi. 1.6 membetulkan nama langkah dari *kalkir* menjadi *kalke* (sinonim *pinjam terjemah*, KBBI), karena *kalkir* adalah kertas jiplak dan bukan istilah linguistik. Versi ini juga menambah enam baris tabel kalke, melarang kalimat penutup berbentuk punchline atau paralelisme retoris, menegaskan bahwa kalimat pendek boleh disambung kata penghubung, menambah larangan *Konsekuensinya* dan *Sedangkan* di awal kalimat, serta menambah aturan register santai, glosarium akronim pada kemunculan pertama, *dapat* pada klaim kemampuan, larangan pola penyangkalan *X, bukan berarti Y*, dan *membahas* sebagai pengganti *mengupas* pada register akademik. Pola penyangkalan kini juga dihitung oleh `scripts/metrik.py`. Seluruhnya lahir dari empat latihan menulis yang dikoreksi langsung oleh penulis pada 24 Oktober 2026: komentar media sosial tentang perubahan iklim, penjelasan santai tentang fuel cell, esai akademik tentang reaktor cepat berpendingin timbal, dan esai akademik tentang mobil listrik. 1.7 menambah pengukuran GPTZero kedelapan dan kesembilan pada model 4.1m, keduanya teks buatan yang hampir atau sama sekali tidak disunting: esai kebijakan tentang rencana PLTN (AI 31 persen) dan esai ensiklopedis tentang fuel cell (AI 4 persen). Keduanya menyimpang jauh dari pola 1 sampai 2 persen human pada pengukuran Agustus. `scripts/metrik.py` naik ke versi 1.4 dengan dua diagnostik baru, yaitu triplet tanpa *seperti* dan kalimat pendek lepas tanpa penghubung. Versi ini juga menambah aturan *saya* untuk esai yang dituturkan penulis sendiri (dicabut untuk register akademik dan