manusiawi

KO | EN

gitlite — search

#python #react #ai #typescript #machine-learning #deep-learning #linux #macos #awesome #cli #awesome-list #pytorch #mcp #openai

(★ 27)

Claude skill that strips AI writing patterns from Malaysian BM text. 32 BM patterns + Indonesian intrusion detection + 24 English patterns.

manusiawi 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Manusiawi

Strips the AI smell from Malaysian Bahasa Melayu writing. You know the tells - "di era globalisasi ini", "secara holistik dan komprehensif", "telah dibuktikan bahawa". Manusiawi catches 32 of these BM patterns, plus 24 English ones for code-switched content, and flags Indonesian vocabulary that leaks in when the model can't tell Malaysian BM from Indonesian.

Claude skill yang buang penulisan berbau AI dari tulisan Bahasa Melayu Malaysia. Korang tahu la gaya tu, "di era globalisasi ini", "secara holistik dan komprehensif", "dengan demikian dapatlah disimpulkan". Manusiawi boleh detect 32 pattern BM macam ni, 24 pattern English untuk content campuran, plus detector perkataan Indonesia yang leak (AI selalu lupa dia tulis untuk orang Malaysia, bukan Indonesia).

## Why I Built This

I got frustrated.

Every time I read AI-generated BM content, I can spot it within 3 seconds. The "era globalisasi" openers, the "holistik" and "komprehensif" vocabulary, the stacked "Selain itu... Di samping itu... Tambahan lagi...". I stop reading the moment I see it. It's that obvious.

The problem is worse in BM than in English. In English, there are tools and guides to strip AI tells - like [blader/humanizer](https://github.com/blader/humanizer), based on Wikipedia's [Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing). In BM? Nothing. No guide, no tool, no skill. LLMs default to karangan-SPM-tahun-90an BM, the most statistically "safe" output that reads like a government pekeliling nobody wants to read.

On top of that, LLMs constantly confuse Malaysian BM with Bahasa Indonesia. You get "kantor" instead of "pejabat", "gimana" instead of "macam mana", "banget" and "dong" sprinkled in - words no Malaysian would ever use. There was no intrusion detection for this either.

So I built Manusiawi:
- **32 BM-specific AI patterns** with before/after examples - the tells that give away AI-written BM
- **Indonesian intrusion detection** - catches Indonesian vocabulary leaking into "BM" output
- **24 English AI patterns** - for the mixed BM-EN content that Malaysians naturally produce when code-switching

## Installation

### Claude Code

Clone directly into Claude Code's skills directory:

```
mkdir -p ~/.claude/skills
git clone https://github.com/abualif120/manusiawi.git ~/.claude/skills/manusiawi
```

### OpenCode

Clone directly into OpenCode's skills directory:

```
mkdir -p ~/.config/opencode/skills
git clone https://github.com/abualif120/manusiawi.git ~/.config/opencode/skills/manusiawi
```

> **Note:** OpenCode also scans `~/.claude/skills/` for compatibility, so a single clone into `~/.claude/skills/manusiawi/` works for both tools.

### Claude Desktop / Claude.ai

Place the `SKILL.md` and `references/` folder in your Claude skills directory.

## Usage

### Claude Code

```
/manusiawi

[paste your BM text here]
```

### Direct Ask

Suruh Claude terus, dalam BM, English, atau campur-campur:

```
Humanize teks ni: [paste text]

Tolong betulkan BM ni, nampak AI sangat: [paste text]

Please humanize this Malay text: [paste text]
```

### Voice Calibration

Nak output ikut gaya tulisan kau sendiri? Bagi sample dulu:

```
/manusiawi

Ni sample tulisan aku untuk voice matching:
[paste 2-3 perenggan tulisan kau]

Sekarang humanize teks ni:
[paste AI text]
```

Skill akan analyse rentak ayat kau, pilihan perkataan, particles yang kau selalu guna dan lepas tu apply pada rewrite, bukan bagi output generic.

## Overview

Based on [Wikipedia's "Signs of AI writing"](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing) guide, maintained by WikiProject AI Cleanup. Adapted from the [blader/humanizer](https://github.com/blader/humanizer) skill with Malaysian BM-specific patterns, Indonesian intrusion detection, and English pattern support for mixed content.

Manusiawi extends the humanizer approach with three additions:

1. **32 BM Malaysia-specific AI patterns** - patterns that only appear when LLMs write BM (era globalisasi openers, holistik/komprehensif vocabulary, telah di- passive overuse, adalah before adjectives, di mana as conjunction).
2. **Indonesian intrusion detection** - catches Indonesian vocabulary that leaks into "BM" LLM output (kantor, mobil, uang, kamu, banget, gue, dong, sih).
3. **24 English AI patterns** - for mixed BM-EN content that Malaysians naturally produce. Keeps technical English, strips AI business/academic/promo speak. See [`references/patterns-en.md`](references/patterns-en.md).

### Key Insight from Wikipedia

> "LLMs use statistical algorithms to guess what should come next. The result tends toward the most statistically likely result that applies to the widest variety of cases."

Dalam konteks BM, LLM default pada bahasa karangan SPM tahun 90an ialah yang paling "selamat", tapi paling tak manusiawi.

## 32 Malaysian BM Patterns Detected (with Before/After)

Pattern numbers match [`references/patterns-bm.md`](references/patterns-bm.md) for full details.

### Content Patterns / Corak Kandungan

| # | Pattern | Before | After |
| --- | --- | --- | --- |
| 1 | **Era globalisasi openers** | "Dalam era globalisasi yang serba mencabar ini, pendidikan memainkan peranan penting..." | "Kos pendidikan tinggi di Malaysia naik purata 6% setahun sejak 2019." |
| 2 | **Significance inflation** | "memainkan peranan yang amat penting, menjadi tunggak utama" | "Sektor IT menyumbang RM44.8 bilion kepada GDP 2024" |
| 3 | **Heavy academic vocab** | "pendekatan holistik yang menyeluruh" | "Kerajaan keluar RM2.3 bilion untuk 14 inisiatif pendidikan" |
| 4 | **Notability inflation** | "telah terkenal seantero negara, mendapat pengiktirafan luas" | "Dah beroperasi sejak 1998. Queue panjang setiap Sabtu pagi." |
| 5 | **Shallow -kan analyses** | "mencerminkan komitmen, menonjolkan kepentingan, menggambarkan keupayaan" | Sebut terus apa yang dibuat |

### Language Patterns / Corak Bahasa

| # | Pattern | Before | After |
| --- | --- | --- | --- |
| 6 | **AI vocabulary** | "ekosistem yang mampan dalam landskap transformatif" | Guna BM biasa - "program ni latih 30,000 guru" |
| 7 | **Promotional language** | "tersergam indah, mempesonakan, mengagumkan" | Fakta, data, nombor konkrit |
| 8 | **Vague attributions** | "para pakar berpendapat, kajian menunjukkan" | "Kajian ILKAP 2023 menemui 67%..." |
| 9 | **Stacked discourse markers** | "Selain itu... Di samping itu... Tambahan lagi... Seterusnya..." | Buang semua, biar fakta cakap sendiri |
| 10 | **Paired phrases** | "cekap dan berkesan, efektif dan efisien" | Pilih satu |
| 11 | **Excessive "telah di-" passive** | "Program telah dilancarkan oleh kerajaan" | "Kerajaan lancarkan program Januari 2024" |
| 12 | **Avoiding "ialah/adalah"** | "JAKIM berperanan sebagai pengesah yang bertindak sebagai..." | "JAKIM ialah pengesah pensijilan halal" |
| 13 | **Rule of three** | "mencipta, membina, dan mengilhamkan generasi inovatif, kreatif, dan berdaya saing" | Bilangan natural, bukan selalu tiga |
| 14 | **Synonym cycling** | "sangat penting, mempunyai kepentingan besar, amat signifikan" | Satu pernyataan specific |

### Style Patterns / Corak Gaya

| # | Pattern | Before | After |
| --- | --- | --- | --- |
| 15 | **Excessive intensifiers** | "sangat membimbangkan, amat kritikal, sungguh serius" | Fakta spesifik - "naik 34% dari Januari ke Jun" |
| 16 | **Development hyperbole** | "mengorak langkah ke hadapan dalam melakar kejayaan" | "Syarikat keluar IPO bulan Mac - target raised RM180 juta" |
| 17 | **Formulaic structure** | Opening kosong → body → optimistic conclusion template | Start dengan fakta paling menarik |
| 25 | **Emoji/bold headers** | 🌟 **Kelebihan:** 💡 **Tip:** ✅ **Manfaat:** | Plain markdown headings |
| 27 | **"Ia adalah" overuse** | "Ia adalah sesuatu yang sangat penting..." | Sebut benda tu terus |
| 28 | **Long-formal sentences** | Ayat 40+ patah perkataan tanpa nafas, gaya pekeliling | Pecah ayat pendek berfokus |

### Grammar Patterns / Corak Tatabahasa

Ni yang most Malaysians don't know - LLMs sebenarnya buat grammar error dalam BM:

| # | Pattern | Before (SALAH) | After (BETUL) |
| --- | --- | --- | --- |
| 18 | **"Adalah" before adjective** | "Adalah penting untuk kita..." | "Penting untuk kita..." |
| 19 | **"Di mana" as conjunction** | "Syarikat di mana saya bekerja..." | "Syarikat tempat saya bekerja..." |
| 20 | **"Yang mana" misuse** | "Sekolah yang mana terletak di KL..." | "Sekolah yang terletak di KL..." |
| 26 | **Literal English translations** | "menjalankan penyelidikan" (do research) | "buat kajian" |
| 29 | **Excessive "untuk" complements** | "mampu untuk menyelesaikan" | "mampu selesaikan" |
| 30 | **Redundant marker pairing** | "Walaupun program baik, tetapi mahal" | "Walaupun program baik, ia mahal" |

### Filler & Hedging / Pengisi & Hedging

| # | Pattern | Before | After |
| --- | --- | --- | --- |
| 21 | **Excessive hedging** | "boleh jadi berkemungkinan akan memberi impak agaknya positif" | "dijangka latih 5,000 belia" |
| 22 | **Filler phrases** | "Perlu diingatkan bahawa, Tidak dapat dinafikan bahawa..." | Buang, langsung pada point |
| 23 | **Vague benefits** | "pelbagai faedah dalam pelbagai aspek dalam pelbagai bidang" | List specific - "(1) peralatan RM3,200, (2) elaun RM500..." |
| 24 | **Overly bright future** | "masa depan yang lebih cerah menanti generasi akan datang" | "Target: 8,000 graduan dalam green tech menjelang 2030" |
| 31 | **"Secara amnya/umumnya"** | "Secara amnya, pelajar Malaysia hadapi cabaran..." | "62% pelajar Tingkatan 5 fail Matematik SPM 2024" |
| 32 | **"Usaha" clichés** | "usaha yang gigih, komitmen yang tinggi, dedikasi tidak berbelah bahagi" | Apa mereka betul-betul buat |

## Indonesian Intrusion Detection

LLM selalu confused antara BM Malaysia dan Bahasa Indonesia sebab akar bahasa sama. Manusiawi detect dan tukar perkataan Indonesia yang bocor masuk. Full list in [`references/indonesian-words.md`](references/indonesian-words.md).

### Nouns / Kata Nama

| Indonesian | BM Malaysia | English |
| --- | --- | --- |
| kantor | pejabat | office |
| mobil | kereta | car |
| uang | wang / duit | money |
| sepeda | basikal | bicycle |
| kulkas | peti sejuk | fridge |
| handuk | tuala | towel |
| bandara | lapangan terbang | airport |
| handphone / HP | telefon bimbit | mobile phone |
| bioskop | panggung wayang | cinema |
| apotek | farmasi | pharmacy |

### Pronouns / Kata Ganti

| Indonesian | BM Malaysia |
| --- | --- |
| kamu (formal) | anda |
| gue / gua | saya |
| lu / loe | awak |

### Verbs & Particles / Kata Kerja & Partikel

| Indonesian | BM Malaysia |
| --- | --- |
| bisa | boleh |
| nggak / gak / enggak | tidak / tak |
| gimana | macam mana / bagaimana |
| mampir | singgah |
| nongkrong | lepak |
| besok | esok |
| kemarin | semalam |
| dong, kok, deh, sih, banget | **buang semua** - no Malaysian equivalent |

### Spelling / Ejaan

| Indonesian | BM Malaysia |
| --- | --- |
| sholat | solat |
| jum'at | jumaat |
| udah | dah / sudah |
| aja | sahaja / je |

"Kereta" warning - Malaysia: kereta = **car**. Indonesia: kereta = **train**. Bila AI tulis "naik kereta ke stesen" dalam konteks public transport, that's Indonesian. BM Malaysia: "naik keretapi ke stesen."

## English Pattern Support

For mixed BM-EN content, Manusiawi also detects 24 English AI patterns - things like "delve", "tapestry", "multifaceted", "leveraging the landscape", sycophantic openings ("Great question!"), and formulaic "The future looks bright" conclusions.

Malaysians code-switch naturally between BM and English. That's fine. What's not fine is AI vocabulary leaking into both languages. The skill keeps technical English (JavaScript, API, MRI) and strips AI English (holistic, groundbreaking, synergize).

Full English pattern list in [`references/patterns-en.md`](references/patterns-en.md).

## Full Example

**Before (AI-sounding BM):**

> Dalam era digital yang semakin berkembang ini, pendidikan online memainkan peranan yang amat penting dalam landskap pembelajaran moden. Pelbagai platform telah dilancarkan di Malaysia bagi menyediakan akses kepada pelbagai kursus yang komprehensif dan holistik. Tidak dapat dinafikan bahawa inisiatif ini akan memberi impak yang positif kepada generasi akan datang.
>
> Selain itu, pelajar mampu untuk mengakses pembelajaran di mana sahaja mereka berada. Di samping itu, pihak industri turut memberikan sokongan yang padu dalam pelbagai aspek. Tambahan lagi, kerajaan telah melaksanakan pelbagai inisiatif yang bersepadu bagi memastikan ekosistem pendidikan yang mampan.
>
> 🌟 **Kelebihan:** Pembelajaran lebih fleksibel
> 💡 **Tip:** Pilih platform yang bereputasi
> ✅ **Manfaat:** Peluang kerjaya yang lebih luas
>
> Secara kesimpulannya, dengan usaha yang gigih dan komitmen yang tinggi daripada semua pihak, masa depan yang lebih cerah menanti generasi akan datang.

**After (humanized):**

> Malaysia ada 4 platform pendidikan online utama: OpenLearning (120,000 pengguna), Taylor's Impact Lab (85,000), Sunway Online (62,000), dan UM Pathway (45,000). Harga kursus antara RM150 untuk modul 4-minggu sampai RM8,500 untuk Sijil Profesional 6 bulan.
>
> Majoriti kursus fokus pada digital marketing, data analytics, dengan project management. Berdasarkan laporan MDEC 2024, 68% pelajar yang habis kursus ni dapat kerja dalam 6 bulan, dengan gaji permulaan antara RM3,200 sampai RM5,800.
>
> Tiga platform terbesar beroperasi secara independent dari universiti konvensional. Satu sahaja, iaitu UM Pathway, yang berada di bawah universiti awam.

## Indonesian Intrusion Example

**Before (contaminated with Indonesian):**

> Gue udah mampir ke kantor kemarin. Gimana sih, pihak HR nggak bisa bantu banget deh. Kamu udah coba belum? Mungkin besok gue harus balik lagi sama bawa handphone buat rekam.

**After (pure BM Malaysia):**

> Saya dah singgah ke pejabat semalam. Macam mana nak cakap, HR tak boleh tolong langsung. Awak dah cuba ke? Mungkin esok saya kena balik lagi, bawa fon untuk rakam.

**What changed:**
- Pronouns: Gue → Saya, Kamu → Awak
- Nouns: kantor → pejabat, handphone → fon
- Verbs: mampir → singgah
- Time: kemarin → semalam, besok → esok
- Particles: banget deh, sih → removed (no Malaysian equivalent)
- Negation: nggak bisa → tak boleh

## References

- [Wikipedia: Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing) - Primary source
- [WikiProject AI Cleanup](https://en.wikipedia.org/wiki/Wikipedia:WikiProject_AI_Cleanup) - Maintaining organization
- [blader/humanizer](https://github.com/blader/humanizer) - Original English humanizer skill that Manusiawi adapts

## Version History

- **1.0.0** - Initial release. 32 BM Malaysia patterns, 24 English patterns, Indonesian intrusion detection, bilingual documentation, MIT license.

## Why "Manusiawi"?

"Manusiawi" in BM means "humanistic" or "human-like." Perkataan yang sesuai untuk skill yang kerjanya buat tulisan AI kedengaran macam manusia tulis.

## Contributing

PR memang welcome. Kalau nak contribute:

- Jumpa pattern AI baru yang skill ni tak tangkap? Tambah je
- Perkataan Indonesia yang LLM selalu leaking? Tambah dalam list
- Contoh before/after yang tak betul atau yang lagi nampak real? Betulkan
- Apa-apa issue pun boleh, termasuk pattern baru, salah detect MY vs ID dll

## License

MIT

manusiawi

manusiawi 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

manusiawi

manusiawi 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README