agentsclimarketplace

Arabic nlp

Skill ArabAgentSkills/Skills/skills/arabic-nlp

Public source-backed Arab-world agent skills: APIs, marketing, sales, support, localization, Arabic NLP, open data, and MENA workflows.

Install
npx -y skills add ArabAgentSkills/Skills --skill arabic-nlp

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 9 stars9 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use this skill when selecting or applying Arabic NLP, tokenization, stemming, diacritization, morphological analysis, stopword, OCR, speech, or language-model tools including CAMeL Tools, Farasa, PyArabic, Tashaphyne, Qalsadi, and Arabic corpora.

SKILL.md

4.0 KB, as published. Nobody here has run it

Arabic NLP

When To Use

Use this skill for arabic-nlp work in Arab/MENA contexts, especially when the prompt includes: Arabic NLP, tokenization, stemming, diacritization, morphology, Farasa, CAMeL, PyArabic.

When Not To Use

  • Do not claim model accuracy without benchmark sources.
  • Do not send private text to third-party APIs without approval.

Required Inputs

  • Country or market.
  • Target vendor, if already selected.
  • Desired workflow: vendor selection, implementation, review/debugging, launch readiness, or source research.
  • Sandbox vs production state.
  • Whether live customer, payment, tax, identity, bank, or payroll data is involved.

Common Workflows

  • Identify the NLP task and deployment constraints.
  • Prefer maintained libraries and official docs or GitHub repos.
  • Separate API services from local libraries.
  • Return evaluation data needs and dialect/classical Arabic caveats.

Default Workflow

  1. Read sources.yml to see available vendors and confidence.
  2. If a vendor is named, read only vendors/<vendor-id>.md for that vendor.
  3. If choosing vendors, compare only vendors in this skill's registry: arabic-stopwords, camel-tools, farasa, pyarabic, qalsadi, tashaphyne.
  4. Load references/integration-checklist.md only for implementation, review, or launch-readiness work.
  5. Use scripts/list-vendors.mjs for a deterministic vendor list when needed.
  6. Answer with source-backed facts, explicit unknowns, and validation steps.

Decision Tree

  • Named vendor: read that vendor file, then answer narrowly.
  • Vendor selection: filter by country, docs access, maturity, and source quality before recommending.
  • Implementation: include auth, sandbox, webhook/callback, retries, idempotency, logging, and error handling only where source-backed.
  • Review/debugging: compare the user's plan or code against the vendor file, sources.yml, and references/integration-checklist.md.
  • Source research: update facts only when an official source, developer portal, GitHub repo, OpenAPI/Postman asset, or government source supports the claim.
  • Missing docs: say Needs vendor access or Unknown from public docs.

Response Contract

  • Start by naming the skill file and vendor/reference files used.
  • Give a short recommendation or implementation path before details.
  • Separate source-backed facts from assumptions and unknowns.
  • Include country/market fit, docs access, docs confidence, and source-quality caveats when selecting vendors.
  • Include a validation checklist with sandbox/test steps, rollback or retry notes, and manual approval gates for high-risk work.
  • Never provide live-action instructions that move money, tax documents, bank data, identity data, payroll data, or outbound messages without explicit human approval.

Files To Read

  • Routing and process: SKILL.md.
  • Vendor facts: vendors/*.md.
  • Source map: sources.yml.
  • Implementation review: references/integration-checklist.md.
  • Example response style: examples/source-backed-answer.md.

Safety Rules

  • Check license and data handling before production use.
  • Avoid sending sensitive Arabic text to external APIs by default.
  • Benchmark on the target dialect and domain.
  • Document tokenizer/stemmer limitations.

Validation Checklist

  • Vendor facts map back to source_urls.
  • Unknowns are labeled instead of guessed.
  • Sandbox and production are separated.
  • Secrets are not printed or committed.
  • High-risk live actions require explicit human approval.
  • Evals in evals/prompts.yml still cover the changed workflow.

Done Criteria

  • The answer names the files read or source-backed references used.
  • The implementation plan includes tests and rollback/verification steps.
  • No unsupported regional, API, compliance, pricing, or endpoint claims are included.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.