Works with the AI Agents you already use
Arabic AI-Tell Gate: Natural Arabic, Same Facts
And when you rewrite by hand or with another prompt, a number changes, a quote gets paraphrased, or «قد» quietly becomes a certainty.
$9
Arabic AI-Tell Gate: Natural Arabic, Same Facts
Example session with this skill installed
This Arabic post was written with ChatGPT and sounds like it. Make it natural MSA, but don't change any facts. (No file attached, so the skill runs its bundled demo: python3 scripts/ar_tell_gate.py demo)
- Read your context and instructions
- Compiled the arabic ai-tell gate
*** DEMONSTRATION on the bundled sample (no input given or input empty). ***
== SCAN: original ==
words 115 | weighted tells 13.5 | density 11.74 per 100 words | verdict FAIL
L1 T01 inflated opener «في عالمنا اليوم» -> Delete the opener; start with the actual point.
L1 T04 role calque «يلعب التعليم الإلكتروني دوراً» -> Use the real verb: what does it do?
L1 T08 signposting filler «ومن الجدير بالذكر» -> Delete the signpost; keep the sentence that follows.
L1 T14 result-chain filler «مما يساهم» -> Split into two sentences and name the result.
L1 T15 adverbial calque «بشكل كبير» -> Cut it, or use a real adverb / number.
L3 T06 buzz metaphor «نقلة نوعية» -> Replace with the literal claim and, if you have it, a number.
L3 T11 not-only-but calque «لا يقتصر الأمر على توفير المحتوى، بل» -> Say both things plainly.
L3 T12 whether-you-are calque «سواء كنت طالباً أو» -> Address the actual reader; drop the audience list.
L3 T16 vague quantity «العديد من» -> Give the number or name the items.
L3 P02 em dash in Arabic prose «—» -> use ، or a new sentence.
L5 T07 journey / exploration frame «دعونا نستكشف» -> Drop the guided-tour frame; state the first fact.
L5 T13 imperative calque «قم ب» -> Use the verb directly.
L5 P01 Latin punctuation in Arabic «,» -> Use ، ؛ ؟ next to Arabic words.
L7 T02 inflated opener «في ظل التطورات المتسارعة» -> Name the specific change or delete.
L7 T06 buzz metaphor «حجر الزاوية» -> Replace with the literal claim.
L7 T10 summary closer «في الختام» -> End on the last real point; cut the recap.
== SCAN: rewrite ==
words 64 | weighted tells 0 | density 0.0 per 100 words | verdict PASS
== MEANING LOCK: LOCKED ==
original carries 1 numbers, 1 Latin tokens, 1 URLs/emails, 1 quotes, 1 negations
all numbers, names, links, quotes, negations and hedges match
(spelled-out numbers and paraphrased claims are not checked - read the rewrite)
RESULT: density 11.74 -> 0.0 (FAIL -> PASS) | lock LOCKED
(Advice text shortened here for length; the script prints it in full.)
Connects securely to your tools. The creator never sees your data.
About this skill
The pain. Arabic from ChatGPT, Gemini or Claude is easy to spot: «في عالمنا اليوم» openers, «يلعب دوراً محورياً», «من الجدير بالذكر», «لا يقتصر الأمر على… بل», «سواء كنت طالباً أو معلماً», «قم بتحميل», a «في الختام» recap, Latin commas and em dashes between Arabic words. English humanizers do not see any of it. And when you rewrite by hand or with another prompt, a number changes, a quote gets paraphrased, or «قد» quietly becomes a certainty.
The outcome. A rewritten Arabic text with a before/after score, a change table (line · tell · before · after), and a meaning-lock verdict that tells you whether every fact survived.
The mechanism.
- A deterministic Python scanner (standard library, read-only, no network) flags 17 MSA phrase tells, Latin punctuation and em dashes next to Arabic, and Egyptian/MSA register mixing, by line number, and scores weighted tells per 100 words: PASS / REVISE / FAIL.
- Your agent rewrites using the bundled rules: fix each finding, add no facts, keep one register.
checkcompares original and rewrite: numbers (Western, Arabic-Indic ٠–٩ and Persian digits folded; 1,250,000 = ١٬٢٥٠٬٠٠٠), Latin-script names, URLs, hashtags and «quotes» must match exactly (BROKEN otherwise); each negation is compared with the two words after it, and hedges (قد، ربما، غالباً) are counted (REVIEW otherwise). The agent loops until it is clean or reports what is open.
Run it with no input and it shows a full labelled demo on a bundled sample.
Limits. Not an AI detector and makes no promise about passing one. Does not translate, fix grammar or check facts against outside sources. Spelled-out numbers and names written in Arabic script are not locked. Register check covers MSA and Egyptian only. The tell list is an editorial list, not a trained model.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse All Claude Skills.
Not an AI detector. No translation, grammar or spelling correction, or fact-checking against outside sources. Spelled-out numbers (ثلاثة آلاف), names in Arabic script and date-format changes are not locked. Register check covers MSA and Egyptian only. Diacritics are removed before matching, so «غيّر» can be read as the negator «غير» and cause a false REVIEW. The tell list is an editorial list with the tool's own thresholds, not a trained or published standard. Input up to 2 MB per file.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 7 days ago
- Software
Python 3.8+Terminal access
Permissions
File Scopes
The script reads files and writes nothing; it opens no network connection. The reference file cites unicode.org and a GitHub page as documentation links only; nothing is fetched.