Skip to content

adelpro/fasaha

v1.1.1MIT

Arabic fluency and correction skill: review and fix AI-generated or AI-translated Arabic so it reads as fluent, native Modern Standard Arabic (MSA) instead of translated/calqued output.

GitHubOpen Website

fasaha (فصاحة) — Arabic Fluency & Correction

Review and fix AI-generated or AI-translated Arabic so it reads as fluent, native Modern Standard Arabic (MSA) instead of translated/calqued output.

Fasaha (فصاحة) means eloquence and correctness of expression in Arabic — text that reads as if it was written by a native speaker, not translated.

Grounded in the QALB (Qatar Arabic Language Bank) annotation guidelines (Zaghouani, Habash & Mohit, CMU-Qatar/Columbia, 2013) plus patterns specific to LLM translation failures.

Direct reference

What it fixes

  • Latin-script leakage — English words left untranslated in technical/marketing copy
  • Sentence-structure calque — Arabic that mirrors English word order and sentence boundaries
  • Word choice / terminology — generic-but-wrong renderings instead of domain-standard terms
  • Morphology, agreement & syntax — gender/number agreement, broken plurals, prepositions, articles
  • Punctuation, hamza & numbers — Arabic punctuation marks, hamza wasl/qat' and medial seat, number rules
  • Dialectal leakage into MSA — dialectal words slipping into formal Arabic

How it works

Running fasaha = correctness pass (Sections 1-7 via the runnable checklist) → register pass (voice-profile.md) → a report of what was flagged and corrected.

It ships living reference files that any agent updates over time:

  • references/checklist.md — the runnable quality gate (every check + a defined output shape)
  • references/terminology.md — accumulating glossary of standard Arabic renderings (web dev, e-commerce)
  • references/llm-failure-log.md — accumulating log of real caught errors (bad → good → section)
  • references/voice-profile.md — self-maintained register file (MSA vs Darija, formality, tone)
  • references/qalb-spelling-rules.md — hamza, punctuation and number rules
  • references/mt-examples.md — worked machine-translation correction examples
  • references/dialect-classification.md — QALB's six-category dialect classification
  • references/sources.md — the QALB source and attribution

Install

Any agent via skills.sh:

npx skills add adelpro/fasaha

Claude Code (skills.sh route):

npx skills add adelpro/fasaha -a claude-code

Claude Code (marketplace route):

claude plugin marketplace add github.com/adelpro/fasaha
claude plugin install fasaha@adelpro-fasaha

Agent Plugins 1.0.0 bundle (plugin.json at repo root) — loadable by compatible clients from Amazon, Cursor, Microsoft, OpenAI (Codex), and Vercel ecosystems.

Usage triggers

  • Translating INTO Arabic
  • Writing original Arabic content
  • Reviewing / correcting / proofreading existing Arabic
  • Any Arabic quality check (e.g. "راجع لي هذا", "صحح العربية", "ترجم هذا للعربية")
  • Proactively before outputting Arabic prose longer than 2-3 sentences

Sources & QALB attribution

Fasaha is grounded in the Qatar Arabic Language Bank (QALB) annotation guidelines — the closest published standard for correcting AI/MT-generated Arabic. The rule sets, worked correction examples, and dialect classification in references/ are condensed and adapted from this source:

Sections 5, 6 and the dialect/spelling references map directly to QALB; Section 7 documents where the 2013 guidelines do not predict modern LLM failure modes.

License

MIT. The skill condenses and adapts rules from the QALB Guidelines v0.90 (Zaghouani, Habash & Mohit). See Sources & QALB attribution and references/sources.md for full attribution.

Author

Adel Ben Yahia