arabic-fix

Fix Arabic text rendering across code, AI agents, and designs.

One library, one CLI, one system prompt. Letter-shaping, BiDi reordering, Unicode normalization, and a 3-section prompt template for any LLM. Arabic flows through terminals, AI agents, log files, and PDFs the way it should.

pip install arabic-fix

from arabic_fix import fix
print(fix("User 42 من الرياض at 09:30"))
# → a properly shaped, correctly ordered, NFC-normalized string

# CLI
cat arabic.txt | arabic-fix
arabic-fix --check file.txt          # CI mode: exit non-zero if changes needed

# AI agent: paste one of the 3 sections from agents/system_prompt.md
#   A · Read Arabic    — for agents that parse / summarize / classify
#   B · Write Arabic   — for agents that emit to terminal / log / email
#   C · Design Arabic  — for HTML / CSS / Tailwind / design tokens

Why

• Letter shaping — ا ب ت (isolated forms) become اتصال (connected) via Unicode Presentation Forms-B.
• BiDi order — User 42 من الرياض ends up with 42 on the visual left and Arabic letters in their connected form, per UAX #9.
• Tashkeel preserved — diacritics survive; مَرْحَبًا round-trips intact.
• Logical CSS only — margin-inline-start, never margin-left.
• Real Arabic fonts — Tajawal, IBM Plex Sans Arabic, Noto Naskh Arabic in priority order; never Inter or Roboto.

What's in the box

Verify

pytest -q                          # 169 tests
mypy arabic_fix                    # strict, clean
python designs/validate_tokens.py  # 73 token schema check

License

MIT — see LICENSE. The brand arabic-fix is locked until v1.0.