agentsclimarketplace
Sign in

Mena customer support

Skill ArabAgentSkills/Skills/skills/mena-customer-support

Use this skill when drafting, reviewing, localizing, or evaluating Arab/MENA customer support macros, Arabic or bilingual support replies, WhatsApp and Instagram support handling, ecommerce payment/refund/delivery/account/subscription/tax responses, complaint de-escalation, escalation rules, and native-review-needed customer-facing Arabic.From its SKILL.md

Install
npx -y skills add ArabAgentSkills/Skills --skill mena-customer-support

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.

SKILL.md

6.5 KB, ~1.3k tokens by cl100k_base, as published. Nobody here has run it

MENA Customer Support

When To Use

Use this skill for Arab/MENA support work involving WhatsApp, Instagram DM/comment, web chat, email, ecommerce orders, payments, refunds, delivery, lost shipments, COD disputes, complaints, account verification, OTP issues, subscriptions, cancellation, and invoice/tax questions.

When Not To Use

  • Do not send customer messages or trigger outbound automation.
  • Do not finalize refunds, cancellations, account closure, identity reset, shipment compensation, or tax corrections.
  • Do not invent merchant policy, country law, platform policy, carrier status, payment status, timelines, or eligibility.

Required Context

  • Country/market, channel, customer language, and public/private context.
  • Scenario and risk level.
  • Merchant policy inputs: refund/return/cancellation terms, carrier status/SLA, payment status, subscription terms, or tax invoice policy.
  • Draft vs approved macro, and sandbox/test queue vs production send state.

Common Workflows

  • Classify scenario, risk, channel, and language.
  • Read the relevant reference before drafting.
  • Draft with acknowledgement, factual status, safe next step, and escalation route.
  • Use placeholders and mark unknowns.
  • Mark dialect, humor, idioms, and public Arabic as native-review-needed.
  • Escalate money movement, identity risk, legal/tax issues, fraud, repeated failure, public anger, or unsafe data requests.

Default Workflow

  1. Read sources.yml, then references/support-tone.md.
  2. Read the scenario file: payment-issue-macros.md, refund-macros.md, delivery-delay-macros.md, complaint-deescalation.md, whatsapp-support.md, or escalation-rules.md.
  3. For implementation or launch readiness, read references/integration-checklist.md.
  4. Draft with placeholders such as [order_id], [case_id], [refund_reference], [carrier_status], [policy_timeline], [secure_link], and [next_followup_at].
  5. If a needed fact is missing, say Unknown from public docs, Needs vendor access, or Needs merchant policy.
  6. Include review flags, escalation, and production gates.

Decision Tree

  • Arabic or bilingual support: default to MSA; use dialect only when country, brand voice, and native review are supplied.
  • WhatsApp: check opt-in/template/service-window assumptions; never ask for OTP, CVV/CVC, full card number, password, or bank credentials.
  • Public social: acknowledge briefly and move private; do not ask for order, phone, address, email, payment, or identity data publicly.
  • Payment/refund: use references and safe last-four only when needed; do not promise refunds, posting dates, or eligibility without policy.
  • Delivery/lost shipment: tie claims to carrier status; do not promise compensation or replacement without policy and carrier review.
  • Account/OTP/identity: never request codes or passwords; escalate suspected takeover, SIM swap, or regulated account issues.
  • Invoice/tax: provide process support only; route legal/tax interpretation to qualified local review.

Response Contract

  • Name files read.
  • State scenario, risk, channel, and review flags.
  • Give the macro/rewrite first, then policy inputs and unknowns.
  • Separate source-backed facts, assumptions, and merchant-policy gaps.
  • Include safe/prohibited data fields and escalation owner.
  • Include sandbox/test and production approval gates for automation or outbound messaging.

Source And Confidence Rules

  • High-confidence claims require public sources or completed review evidence.
  • Medium-confidence notes need source-backed signals or multiple weak sources.
  • Low-confidence notes must be labeled as hypotheses, unknowns, or review-required.
  • Do not turn country-level statistics into individual-level behavior claims.

Anti-Stereotype Rules

  • Never assign traits to nationality, religion, ethnicity, dialect group, gender, age, or income group.
  • Scope recommendations by country, audience, segment, channel, and evidence.
  • Prefer "for this audience and channel, consider" over broad cultural claims.

Output Rules

  • Give practical drafts, plans, templates, checklists, or QA steps.
  • Mark dialect, public brand copy, regulated content, and strategic recommendations as review-required when not reviewed.
  • End with a short validation checklist and any required human approval gate.

Files To Read

  • Sources: sources.yml.
  • Tone: references/support-tone.md.
  • WhatsApp/social: references/whatsapp-support.md.
  • Refunds: references/refund-macros.md.
  • Payments/account/OTP: references/payment-issue-macros.md.
  • Delivery/lost shipment/COD: references/delivery-delay-macros.md.
  • Complaints: references/complaint-deescalation.md.
  • Escalation: references/escalation-rules.md.
  • Examples: examples/support-macros.md.

Safety Rules

  • Never request OTP, password, CVV/CVC, full card number, bank credentials, private keys, or unnecessary identity documents.
  • Do not promise refunds, compensation, legal rights, delivery dates, bank posting timelines, cancellation success, or tax outcomes without supplied policy/review.
  • Move payment, identity, invoice, phone, address, and order details to private or secure channels.
  • Treat payments, refunds, chargebacks, account access, identity, tax, high-value lost shipments, and public complaints as high-risk.
  • Separate sandbox/test queues from production sends; require explicit approval for live outbound messages or customer-record changes.
  • Keep retries idempotent when automating ticket updates or message sends.

Validation Checklist

  • Policy inputs are present or marked Needs merchant policy.
  • Unsupported facts are marked Unknown from public docs or Needs vendor access.
  • Public replies contain no private customer data.
  • Dialect, humor, idioms, and public Arabic are flagged native-review-needed.
  • High-risk scenarios include human escalation and no live-action instruction.

Done Criteria

  • The answer names the files read.
  • The macro uses safe placeholders and avoids prohibited data collection.
  • Review flags and escalation rules are explicit.
  • Missing policy, platform, legal, carrier, or gateway facts are labeled instead of guessed.
  • Production use includes sandbox/test validation and explicit approval gates.

What ships with it: 16 files

42.1 KB alongside SKILL.md, 1 of them executable

evals/

examples/

references/

scripts/

vendors/

Gives 0 of the 12 instructions most customer support skills give in ~1.3k tokens

Counted across 123 of the 124 authors here whose files we hold, read 2026-08-07

  • Call RUBE_SEARCH_TOOLS first to get current schemasin 12 of 123, across 4 files
  • Confirm connection status is ACTIVE before running workflowsin 12 of 123, across 4 files
  • Stop and ask for clarification if required inputs are missingin 9 of 123, across 2 files
  • Call RUBE_MANAGE_CONNECTIONS with the helpdesk toolkitin 9 of 123, across 2 files
  • Use both timestamp and ID for cursor navigationin 8 of 123, across 1 file
  • Implement backoff on 429 responsesin 8 of 123, across 1 file
  • Parse response data defensively with fallback patternsin 8 of 123, across 1 file
  • Use this skill only when the task clearly matches the scopein 8 of 123, across 1 file
  • Pass a JSON file as the positional argumentin 7 of 123, across 1 file
  • Specify output format with the --format flagin 7 of 123, across 1 file
  • Run health, churn, and expansion scripts togetherin 7 of 123, across 1 file
  • Verify output files contain expected records before continuingin 7 of 123, across 1 file

Said here and by no other author read

  • read sources and tone references before drafting
  • classify scenario risk channel and language
  • draft with acknowledgement status next step and escalation
  • use placeholders for unknown values
  • mark missing facts as unknown or needing access
  • name files read in the response

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. 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.