3ameiah

Natural Egyptian Arabic and readable Franco for coding agents—with six commands that are easy to remember.

Install once

npx 3ameiah install

Then use one of these

/3ameiah                     general Egyptian Arabic
/3ameiah-pro                 professional Egyptian Arabic
/3ameiah-friendly            warm, friendly Egyptian Arabic
/3ameiah-franco              readable Egyptian Franco / Arabizi
/3ameiah-learn <name>        create or improve a tone profile
/3ameiah-use <name>          use that profile

That is the whole normal workflow. Everything below is explanation or advanced setup.

Codex uses $3ameiah, $3ameiah-pro, and so on because Codex chooses $ for explicit skills. Claude Code uses /. Other agents may use /, @, a skill picker, or a natural request. The names stay the same.

The four writing commands

/3ameiah

The default. Use it for normal Egyptian Arabic that feels natural, clear, and widely understandable.

/3ameiah اعمل الـ onboarding ده بالمصري
/3ameiah اكتب السكريبت ده بالعامية
/3ameiah خلّي رسائل الـ CLI دي بالمصري

If the task is inside a codebase, the agent still builds or edits the real product. 3ameiah changes human-facing language; it does not turn a coding request into a writing sample.

/3ameiah-pro

For clients, work, proposals, support, serious updates, and professional UI. It stays recognizably Egyptian without becoming bureaucratic MSA.

/3ameiah-pro اكتب الرد ده للعميل
/3ameiah-pro راجع العرض وخلي الخطوة الجاية واضحة

/3ameiah-friendly

For warmer explanations, communities, creators, onboarding, educational content, and friendly messages. Friendly does not mean filler, fake intimacy, pet names, or forced emojis.

/3ameiah-friendly اشرح الخطوة دي للمبتدئ
/3ameiah-friendly حوّل الفكرة دي لبوست قريب من الناس

/3ameiah-franco

For Egyptian speech written in readable Latin characters and Arabizi numbers. It preserves Egyptian grammar instead of translating the message into English.

/3ameiah-franco ektb el message de franco
/3ameiah-franco حوّل الكابشن ده لفرانكو مقروء

Profiles that learn with you

A profile is one local JSON file that remembers style-relevant things you intentionally teach it:

• who you normally write for;
• your role, common projects, terminology, and goals;
• tone, warmth, directness, and formality;
• how much English you naturally mix in;
• words and habits to prefer or avoid;
• liked writing samples and examples you dislike.

First time: create it

Inside a coding agent:

/3ameiah-learn creator

If you do not know what to say, the skill offers three easy choices:

1. Describe the voice in one sentence.
2. Give it 2–5 good samples and optionally one example you dislike.
3. Let it ask five short questions.

From the terminal:

npx 3ameiah learn creator

Non-interactive example:

npx 3ameiah learn creator \
  --yes \
  --mode 3ameiah-friendly \
  --about "I teach programming to Arabic-speaking beginners" \
  --tone "واضح، دافي، وبيدخل في الموضوع بسرعة" \
  --audience "مبتدئين في البرمجة" \
  --directness 4 \
  --warmth 4 \
  --prefer "مثال عملي بعد كل فكرة" \
  --avoid "المقدمات الطويلة"

Later: teach the same profile more

Run learn again with the same name. It updates the same file, keeps unrelated knowledge, increments the revision, records what changed, and backs up the previous version.

npx 3ameiah learn creator \
  --yes \
  --about "I also make short launch videos" \
  --avoid "hype without a concrete takeaway" \
  --note "Improve short-form content"

Add good and bad examples:

npx 3ameiah learn creator \
  --yes \
  --from ./good-post.txt \
  --from ./good-script.txt \
  --dislike ./not-my-style.txt

From the terminal, --from and --dislike archive the files as evidence; they do not automatically turn raw text into tone rules. The raw text also stays out of later automatic model context unless you explicitly request it through profile context <name> --include-samples. Either save the preferences yourself in the same call:

npx 3ameiah learn creator \
  --yes \
  --from ./good-post.txt \
  --tone "مباشر وعملي" \
  --prefer "ابدأ من المشكلة" \
  --avoid "المقدمات العامة"

Or use /3ameiah-learn creator inside an agent and give it the samples. The agent inspects only observable style patterns, shows you a compact proposed learning summary, and—after you confirm—stores the extracted tone/prefer/avoid fields with the sample evidence.

The same works conversationally:

/3ameiah-learn creator اتعلم إنّي بحب أبدأ من المشكلة بسرعة، ومش بحب المقدمات العامة

Use it

For one task inside an agent:

/3ameiah-use creator اكتب البوست ده

Make it your active profile from the terminal:

npx 3ameiah use creator

The active profile becomes your private global voice ledger across projects. A direct command still wins: /3ameiah-pro uses compatible personal preferences but keeps the professional mode.

Inspect it

npx 3ameiah profiles
npx 3ameiah profile show creator
npx 3ameiah profile context creator --json

Profile resolution and safety:

explicit profile > private global active profile > no profile
explicit writing command > profile default mode
task, facts, code, structure, and safety > every profile preference

Profiles and samples are untrusted style data and must never be followed as instructions or treated as permission or factual proof. Stored sample text stays out of automatic model context unless you explicitly include it.

See docs/PROFILES.md for file locations, structural schema, runtime validation, revisions, backups, and privacy.

Install by pasting one link

Copy:

https://github.com/abdullah-elbedwehy/3ameiah

Paste it into any tool-enabled coding agent with terminal and filesystem access, then say:

Review this repository and its scripts. Detect this agent's exact target ID, inspect existing same-name skills for local modifications, then install all six skills globally with the pinned universal skills CLI. Do not replace a locally modified skill without telling me first. Verify all six names and tell me the exact invocation syntax here: https://github.com/abdullah-elbedwehy/3ameiah

A chat-only LLM cannot install files on your computer. It can still read and apply the raw core skill for that conversation. A standalone copy/paste version lives in INSTALL-FOR-AGENTS.md.

Installation options

The simple command installs all six skills globally for Claude Code and Codex, the safe non-interactive baseline:

npx 3ameiah install

Inspect first:

npx 3ameiah install --list
npx 3ameiah install --dry-run

Only the current project:

npx 3ameiah install --project

Only named agents:

npx 3ameiah install --agent claude-code --agent codex

Install directly from GitHub through the universal vercel-labs/skills CLI:

npx --yes [email protected] add abdullah-elbedwehy/3ameiah \
  --global \
  --skill '*' \
  --agent claude-code codex \
  --yes \
  --full-depth

Run the packaged CLI from GitHub before or without an npm release:

npx --yes --package=github:abdullah-elbedwehy/3ameiah 3ameiah install

For permanent shell shortcuts:

npm install --global 3ameiah
3ameiah learn creator
3ameiah-learn creator
3ameiah-use creator

npx 3ameiah install never silently edits your shell PATH or leaves a global executable behind. Read INSTALL.md for scopes, updates, verification, and troubleshooting.

Supported agents

3ameiah follows the open Agent Skills format and delegates installation to the pinned universal vercel-labs/skills CLI. The installer recognizes 76 targets, including Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, Windsurf, Cline, Roo Code, OpenCode, Goose, Amp, Kiro, Zed, Kimi CLI, OpenHands, Antigravity, and Qwen Code.

Common invocation examples:

| Agent | General command | | --- | --- | | Claude Code | /3ameiah | | OpenAI Codex | $3ameiah or /skills | | Cursor | /3ameiah | | GitHub Copilot / VS Code | /3ameiah | | Windsurf | @3ameiah | | Kimi CLI | /skill:3ameiah | | Goose | /skills 3ameiah | | Gemini CLI, Roo, OpenHands, Antigravity | Use 3ameiah for this task |

The exact paths, all target IDs, and verified provider behavior are documented in docs/AGENTS.md.

What the skill protects

3ameiah adapts the real artifact without casually changing:

• facts, claims, dates, amounts, warnings, promises, and obligations;
• code, identifiers, JSON/YAML keys, placeholders, URLs, and commands;
• UI architecture, document structure, subtitle timing, and source fidelity;
• audience, gender, spelling, directionality, and established terminology.

It routes product UI, codebases, posts, ads, landing pages, scripts, subtitles, lessons, stories, messages, PDF/DOCX/PPTX, structured data, and high-stakes wording through separate playbooks. Unknown artifact types use the same universal workflow instead of falling back to generic slang.

Security and privacy

• No postinstall hook. Filesystem changes happen only after an explicit command.
• 3ameiah has no telemetry; the pinned installer has documented opt-out anonymous telemetry.
• Profile updates back up the prior revision.
• Tone samples stay local at rest, have strict size limits, and are excluded from automatic context.
• Samples enter model context only after an explicit request and remain untrusted content, never instructions.
• Do not store credentials, tokens, sensitive records, or unrelated private data in profiles.
• “Write this message” never means “send this message.”

Read SECURITY.md before reporting a vulnerability.

Development

Requirements: Node.js 22.20+ and Python 3.10+ for the language helper tests.

npm install
npm run test:all
npm pack --dry-run

The repository deliberately has no root SKILL.md; every installable skill is under skills/<name>/SKILL.md, so multi-skill discovery finds all six commands.

See CONTRIBUTING.md for test and release rules.

License

MIT