Community라이팅 & 에디팅github.com

arabic-code-review

Performs code reviews and writes them in clear, respectful Arabic for Arab engineering teams - structured summary, severity labels in Arabic with English in parentheses, technical terms glossary, Markdown that survives right-to-left rendering on GitHub/GitLab, and checks for Arabic-specific bugs (normalisation for search, grapheme length, collation, RTL UI). Use when the user asks for a review in Arabic, when a team's working language is Arabic, when writing pull request comments for Arabic-speaking developers, or when the user says «راجع الكود»، «مراجعة الشيفرة بالعربي»، «اكتب ملاحظات المراجعة بالعربية»، «Code review بالعربي».

arabic-code-review란 무엇인가요?

arabic-code-review is a Claude Code agent skill that performs code reviews and writes them in clear, respectful Arabic for Arab engineering teams - structured summary, severity labels in Arabic with English in parentheses, technical terms glossary, Markdown that survives right-to-left rendering on GitHub/GitLab, and checks for Arabic-specific bugs (normalisation for search, grapheme length, collation, RTL UI). Use when the user asks for a review in Arabic, when a team's working language is Arabic, when writing pull request comments for Arabic-speaking developers, or when the user says «راجع الكود»، «مراجعة الشيفرة بالعربي»، «اكتب ملاحظات المراجعة بالعربية»، «Code review بالعربي».

지원 대상~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/MahsumahCloud/saudi-claude-skills/tree/HEAD/skills/arabic-code-review

Installed? Explore more 라이팅 & 에디팅 skills: steipete/notion, langchain-ai/langchain, bytedance/podcast-generation · View all 6 →

즐겨 사용하는 AI에게 물어보기

이 에이전트 스킬이 미리 로드된 새 채팅을 엽니다.

문서

Arabic Code Review

Arabic version: SKILL.ar.md. Answer in the language the user writes in.

Reviewing code in Arabic involves more than translating English review comments. Engineering vocabulary is mixed, identifiers are English, and Markdown renders right to left on most platforms when a comment starts with an Arabic letter. A good Arabic review is technically exact, easy to act on, kind in tone, and formatted so that code and English terms stay readable inside Arabic sentences.

When to use

  • The user asks for a code review "in Arabic" or writes the request in Arabic.
  • Writing pull/merge request comments for a team whose working language is Arabic.
  • Mentoring junior developers in Arabic, where the explanation matters as much as the verdict.
  • Reviewing code in Arabic-language products, where Arabic-specific bugs are likely.

When NOT to use

  • The team reviews in English. Use the normal review flow, though the Arabic-specific bug checks below still apply to Arabic products.
  • Pure UI direction or copy questions. Use arabic-rtl-ui or arabic-copywriting.

Method

  1. Understand the change first. Read the description, linked issue and the full diff. Identify the intent in one sentence before judging lines.
  2. Review in passes: correctness and security, then data and concurrency, then performance, then readability and tests, then Arabic/locale specifics.
  3. Classify every finding with one severity label (table below). One finding, one comment. Do not bundle unrelated points.
  4. Write each comment in a fixed shape: what (the observation, pointing to the exact line) → why (the risk or cost) → suggestion (code or concrete step). End with a question when you are not sure.
  5. Use terms consistently: Arabic term with the English in parentheses on first use in the review, then the Arabic term alone. Identifiers, file names, commands and error messages stay in English inside backticks.
  6. Format for RTL rendering (rules below) and preview if the platform supports it.
  7. Write the summary last, at the top: verdict, blocking count, and the one or two most important points. Include genuine praise for something specific.

Severity labels

LabelMeaningMerge?
حرج (Blocker)Bug, security hole, data loss, broken build or contractMust fix before merge
مهم (Major)Likely defect, missing test for risky logic, notable performance or maintainability costShould fix before merge, or agree on a follow-up
اقتراح (Suggestion)Better approach, clearer naming, simplificationAuthor decides
ملاحظة صغيرة (Nit)Style, typo, formatting not caught by toolingOptional
إشادة (Praise)Something done well, named specifically-

Tone

  • Talk about the code, not the person: «هذا السطر قد يُرجع undefined عندما...» instead of «أنت أخطأت هنا».
  • Explain the reason. «لأن» turns an order into a lesson.
  • Offer a fix or a direction, not only the problem.
  • Ask when unsure: «هل المقصود أن تُستدعى هذه الدالة مرة واحدة فقط؟»
  • Avoid absolute words («دائمًا»، «أبدًا»، «كارثي») unless they are literally true.
  • Avoid sarcasm and rhetorical questions. They read harsher in text than in person.
  • Use «نحن» for shared ownership when suggesting conventions: «لنعتمد...»

Formatting for right-to-left rendering

GitHub and GitLab set paragraph direction from the first strong character. A comment that starts in Arabic is laid out right to left, and English fragments inside it can reorder.

  • Put every identifier, path, command, value and English error in backticks: getUserById, src/api/users.ts, npm test. Inline code stays LTR and isolates well.
  • Put multi-line code in fenced blocks with a language tag. Code blocks always render LTR.
  • Do not start a paragraph or list item with an English word or a backtick span. Begin with an Arabic word, so the line direction is predictable: «الدالة parse لا...» rather than «parse لا...».
  • Avoid ending a sentence with an English word followed by punctuation. Rephrase, or put the term in backticks.
  • Numbered lists: the numbers render on the right in RTL paragraphs. Keep items short and start each with Arabic text.
  • Suggestion blocks (GitHub ```suggestion) contain code only, with no Arabic inside.
  • Use Arabic punctuation «، ؟ ؛» in prose. Keep code punctuation inside code.

Terminology glossary

EnglishArabic
Code reviewمراجعة الشيفرة
Pull requestطلب دمج
Refactorإعادة هيكلة
Bugعلّة / خلل
Regressionتراجع وظيفي
Edge caseحالة حدّية
Race conditionحالة تسابق
Deadlockاستعصاء / جمود متبادل
Null pointer / null referenceمرجع فارغ
Memory leakتسرّب الذاكرة
Dependency injectionحقن الاعتماديات
Dependencyاعتمادية
Side effectأثر جانبي
Immutableغير قابل للتغيير
Idempotentمتساوي الأثر عند التكرار
Concurrencyالتزامن
Asynchronousغير متزامن
Callbackدالة استدعاء راجع
Exception handlingمعالجة الاستثناءات
Input validationالتحقق من المدخلات
Sanitisation / escapingتعقيم المدخلات / تهريب المحارف
SQL injectionحقن SQL
Cross-site scripting (XSS)البرمجة عبر المواقع (XSS)
Authenticationالمصادقة
Authorisationالتفويض / الصلاحيات
Secret / credentialسرّ / بيانات اعتماد
Cachingالتخزين المؤقت
Technical debtالدَّين التقني
Unit testاختبار وحدة
Integration testاختبار تكامل
Test coverageتغطية الاختبارات
Lintingالفحص الساكن للأسلوب
Backward compatibilityالتوافق مع الإصدارات السابقة
Breaking changeتغيير كاسر
Loggingالتسجيل (السجلات)
Performance bottleneckعنق زجاجة في الأداء

These are common renderings, not an official standard. Agree on the team's terms once and keep them.

Arabic-specific bugs to look for

  • Search without normalisation. Users type «احمد» and expect «أحمد». Normalise both the stored text and the query: alef forms (أ إ آ ٱ → ا), alef maqsura (ى → ي), ta marbuta (ة → ه, if the product accepts this), tashkeel removal (U+064B-U+065F, U+0670), tatweel removal (U+0640), and Arabic-Indic/Persian digits → ASCII. Store the normalised form in a separate column or index field, never instead of the original.
  • Length checks with .length. «مُحَمَّد» has 8 UTF-16 code units but 4 user-perceived characters (graphemes). Use Intl.Segmenter (granularity grapheme) for limits shown to users. Use bytes only for storage limits.
  • Sorting with default sort(). Use Intl.Collator('ar') or the database's Arabic collation. Code-point order puts some letters, such as hamza forms, in surprising places.
  • Case functions. toUpperCase()/toLowerCase() do nothing to Arabic. Case-insensitive comparisons that rely on them silently fail to normalise Arabic.
  • Regex \w and \b. In JavaScript they are ASCII-only, even with the u flag, so they never match Arabic letters. Validation such as /^\w+$/ for names rejects Arabic names. Use /^[\p{L}\p{M} ]+$/u (letters plus combining marks, which covers tashkeel) or a script-specific \p{Script=Arabic}.
  • Hard-coded direction. left/right in styles, unmirrored arrows, phone numbers not isolated. Point to arabic-rtl-ui.
  • String concatenation for translations. Breaks word order and agreement. Ask for ICU messages.
  • Number parsing. parseInt('٤٥') returns NaN. Normalise digits before parsing user input.

Reference normaliser for review suggestions:

export function normalizeArabicForSearch(s) {
  return s
    .normalize("NFC")
    .replace(/[ً-ٰٟ]/g, "")   // tashkeel
    .replace(/ـ/g, "")                  // tatweel
    .replace(/[آأإٱ]/g, "ا") // آ أ إ ٱ -> ا
    .replace(/ى/g, "ي")            // ى -> ي
    .replace(/ة/g, "ه")            // ة -> ه (only if your product accepts it)
    .replace(/[٠-٩]/g, d => String(d.charCodeAt(0) - 0x660))
    .replace(/[۰-۹]/g, d => String(d.charCodeAt(0) - 0x6F0));
}

Checklist

  • Intent of the change stated in one sentence.
  • Every finding has exactly one severity label.
  • Each comment has what → why → suggestion.
  • Technical terms: Arabic first, English in parentheses on first use.
  • Identifiers, paths, commands and errors in backticks. Code in fenced blocks.
  • No paragraph starts with an English token.
  • Tone is about the code, not the person. No sarcasm.
  • Security, data and concurrency checked before style.
  • Arabic-specific checks done (normalisation, graphemes, collation, regex, RTL, digits).
  • Summary at the top with verdict, blocker count and specific praise.

Do / Don't

DoDon't
«هذا الاستعلام يدمج مدخلات المستخدم مباشرة، مما يسمح بحقن SQL.»«الكود هذا سيء.»
Start lines in Arabic, put identifiers in backticksStart a list item with userId and let bidi scramble the line
Explain why and propose codeLeave "fix this" without context
Use one Arabic term per conceptAlternate «علّة»، «باق»، «خطأ برمجي» in one review
Mark nits as nitsBlock a merge for naming preferences
Praise a specific decisionWrite empty praise such as «عمل رائع» with nothing specific

Examples

Diff under review

// routes/users.js
router.get("/users/search", (req, res) => {
  const name = req.query.name;
  const rows = db.query(`SELECT * FROM users WHERE name LIKE '%${name}%'`);
  res.json(rows.filter(r => r.name.length <= 20));
});

Review written in Arabic

الخلاصة: التغيير يضيف بحثًا عن المستخدمين بالاسم، وفكرته واضحة. لكن فيه ملاحظتان حرجتان تمنعان الدمج الآن: «حقن SQL (SQL injection)» وعدم انتظار نتيجة الاستعلام. وهناك اقتراح يخص البحث بالعربية.

حرج (Blocker): حقن SQL السطر 4 يدمج قيمة req.query.name داخل نص الاستعلام مباشرة. يستطيع أي مستخدم إرسال ' OR '1'='1 فيقرأ جدول المستخدمين كاملًا. الحل استعلام بمعاملات (parameterized query):

const rows = await db.query(
  "SELECT id, name FROM users WHERE name LIKE $1",
  [`%${name}%`]
);

حرج (Blocker): نتيجة غير منتظرة الدالة db.query غير متزامنة (asynchronous) في هذه المكتبة، والشيفرة لا تستخدم await، فتصبح rows وعدًا (Promise) لا مصفوفة، وسيفشل filter. اجعل المعالج async وانتظر النتيجة كما في المثال أعلاه.

مهم (Major): إرجاع كل الأعمدة الاستعلام SELECT * يعيد أعمدة قد تكون حساسة، مثل بصمة كلمة المرور. نقترح تحديد الأعمدة المطلوبة فقط.

اقتراح (Suggestion): البحث بالعربية لو كتب المستخدم «احمد» فلن يجد «أحمد». نقترح حفظ نسخة مطبّعة من الاسم في عمود مستقل والبحث فيها بعد تطبيع الاستعلام بالدالة normalizeArabicForSearch.

ملاحظة صغيرة (Nit): الشرط r.name.length <= 20 يعدّ وحدات UTF-16 لا الحروف الظاهرة. إن كان الحد معروضًا للمستخدم فاستخدم Intl.Segmenter.

إشادة (Praise): تسمية المسار /users/search واضحة ومتسقة مع بقية المسارات.

Template

**الخلاصة:** <وصف التغيير في جملة> — <القرار: جاهز للدمج / يحتاج تعديلات> — حرج: <عدد>، مهم: <عدد>.

**حرج (Blocker): <عنوان قصير>**
<ما الملاحظة، مع رقم السطر> <لماذا هي خطرة> <المقترح أو الشيفرة>

**مهم (Major): <عنوان>**
...

**اقتراح (Suggestion): <عنوان>**
...

**ملاحظة صغيرة (Nit):** ...

**إشادة (Praise):** <شيء محدد أُحسن صنعه>

References

Individual skills in this repo

This repo contains 2 individual skills — each has its own dedicated page.

관련 스킬