Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Arabic AI-Tell Gate: Score, Rewrite & Fact-Lock

    1

    Most "humanizer" skills are written for English and swap synonyms, so the English skeleton stays.

    $9

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Arabic AI-Tell Gate: Score, Rewrite & Fact-Lock

    Arabic AI-Tell Gate: Score, Rewrite & Fact-Lock

    Example session with this skill installed

    Show me what this does on the bundled sample before I use it on my own Arabic post: score it, rewrite it in native Arabic, and prove no figure, link or the ministry's name was lost.

    • Read your context and instructions
    • Compiled the arabic ai-tell gate

    Ran the bundled demo flow (scan → rewrite → lock). Real output, unedited excerpt:

    ──────────────────────────────────────────────────────────────────────────────
     ARABIC AI-TELL GATE | SCAN   sample: AI draft   (112 words)
    ──────────────────────────────────────────────────────────────────────────────
     SCORE 0/100   HEAVY AI TELLS   hits=25  weighted points=45
     HOW THAT SCORE WAS REACHED
       45 weighted points x 100 / 112 words = 40.18 per 100 words
       100 - 6 x 40.18 = -141.1 -> 0 (rounded half up, floor 0)
       bands: 85-100 NATURAL | 60-84 REVISE | 0-59 HEAVY AI TELLS
     BY FAMILY
       T1 stock frames                 3 hit(s)    8 pt
       T2 empty emphasis               2 hit(s)    4 pt
       T3 inflated phrases             6 hit(s)   12 pt
       T4 structural calques           6 hit(s)    9 pt
       T5 chatbot residue              1 hit(s)    3 pt
       T6 stacked connectors           3 hit(s)    5 pt
       T7 punctuation/orthography      4 hit(s)    4 pt
     RHYTHM  7 sentences | mean 16 words | CV 0.33 (flag below 0.25)
     NOTE  1 quoted span(s) not scored for T1-T6 (quotes must stay verbatim)
    ──────────────────────────────────────────────────────────────────────────────
     T7  text  two digit systems in one text (0-9 and ٠-٩)  (+1)
                evidence: 0-9 x10 | ٠-٩ x7 digits
                instead:  pick one, keep it throughout
     T1  L1    opener 'in today's world'  (+3)
                evidence: في عالم اليوم المتسارع، يلعب الذكاء الاصطناعي دورا…
                instead:  start with the fact itself
       …(23 more findings, same format)…
    
    ──────────────────────────────────────────────────────────────────────────────
     ARABIC AI-TELL GATE | FACT LOCK   original: AI draft   rewrite: native rewrite
    ──────────────────────────────────────────────────────────────────────────────
     FACTS LOCKED   errors=0
     KEPT / IN ORIGINAL
       numbers            5/5
       URLs               1/1
       Latin names/terms  1/1
       quoted texts       1/1
       protected terms    1/1
       negation words     4 -> 1
       percent markers    1 -> 1
    ──────────────────────────────────────────────────────────────────────────────
     REVIEW number order: 62 1200 2025 340 2030 -> 1200 62 2025 340 2030
            same numbers in a different order: confirm each still belongs to its noun
     REVIEW negation: 4 -> 1 | no longer negated: تتردد, شك, مجرد
            negation changed: confirm no claim flipped
     Not checked: Arabic names unless passed with --keep, spelled-out numbers,
     and meaning changes that keep the same tokens. Read REVIEW items yourself.
    ──────────────────────────────────────────────────────────────────────────────
    
     BEFORE 0/100 HEAVY AI TELLS  ->  AFTER 100/100 NATURAL  |  FACTS LOCKED
    

    Result: 0/100 HEAVY AI TELLS → 100/100 NATURAL, FACTS LOCKED. The two REVIEW lines were checked: the numbers moved because the sentence now opens on the teachers; the dropped negations were the filler phrases لا شك / ليس مجرد / لا تترددوا, so no claim flipped.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    The problem

    AI-written Arabic is grammatical and still reads as translated. It opens with في عالم اليوم المتسارع, says قامت الوزارة بإطلاق where an Arabic writer says أطلقت الوزارة, starts every sentence with علاوة على ذلك, uses Latin commas, and closes with في الختام. Most "humanizer" skills are written for English and swap synonyms, so the English skeleton stays. Rewrites also quietly drop facts: a figure, a link, a name.

    What you get

    A native Modern Standard Arabic rewrite, a measured before/after score, and a fact-lock report that shows every number, date, link, Latin name, quote and protected Arabic name survived.

    How it works

    1. Scan (deterministic). A Python script measures 8 families: stock frames, empty emphasis, inflated metaphors (يلعب دوراً محورياً، حجر الزاوية، ليس مجرد… بل), English-shaped syntax (قام بـ، يتم + مصدر، من قبل، حيث أن), chatbot residue, sentence-opening connectors, Latin punctuation and digit mixing, and uniform rhythm. The score prints its arithmetic.
    2. Clean (deterministic). Safe punctuation fixes into a new file: ، ؛ ؟ « », stretching tatweel. Code, URLs and the prefix بـ/لـ stay untouched.
    3. Rewrite (edited judgement). Verb first, the doer as subject, the concrete claim instead of the metaphor, guided by a bundled reference that lists the native replacement for every pattern.
    4. Fact lock. Compares original and rewrite: numbers across 0-9 / ٠-٩ / ۰-۹ (counted, sign and leading zeros kept), dates and times, URLs, emails, handles, Latin names, quotes, and the Arabic names you pass with --keep. Negation, number-order and percent changes are flagged for review. It exits 1 if anything was lost.

    Limits

    It is not an AI detector and never claims "undetectable". It does not proofread إعراب, translate from scratch, or check Arabic names you did not pass to --keep. The catalogue is tuned for MSA; dialect rewrites are supported but not scored for dialect naturalness.

    Built by a native Arabic speaker who works in Arabic linguistics and verification tooling. Standard-library Python, no network, no API keys.

    Not an AI-authorship detector. Does not proofread grammar (إعراب) or spelling, and does not translate from scratch. Arabic names are checked only when passed with --keep; spelled-out numbers (عشرة) are not matched to digits; meaning changes that keep the same tokens (most → all) need human review. Pattern catalogue is tuned for Modern Standard Arabic; some flagged forms are legitimate in context (listed in the bundled reference). Plain .txt/.md input, UTF-8.

    How to use

    1. 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. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    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

    Requires

    Python 3.8+
    Terminal / shell
    Python

    Permissions

    File Scopes

    text
    files
    you
    name
    (read)
    the
    new
    OUT
    file
    given
    to
    clean
    (write)

    Reads only the files you name. clean writes only the new output file and refuses to overwrite anything. No network calls.

    Listed7 days ago

    What's inside

    Frequently Asked Questions

    No. It measures named writing patterns and shows the arithmetic. A clean score does not prove a human wrote a text, and a low score does not prove AI did. The report never claims a text is 'undetectable'.

    The lock script compares the original and the rewrite and exits 1 if any number, date, URL, email, handle, Latin name, quote or protected Arabic name is missing. Negation, number-order and percent changes are listed as REVIEW items for you to confirm.

    The rewrite can target a dialect you name, and the fact lock works for any register. The tell catalogue is tuned for Modern Standard Arabic, so a dialect text is not scored for dialect naturalness; the report says so.

    Python 3.8+ and a terminal. Standard library only, no network, no API keys. Without input it runs a full labelled demonstration.

    Pass them with --keep "name one,name two". They are matched as whole words with common prefixes (و ف بـ لـ كـ ال), with أ إ آ treated as ا, and without merging ى and ي, so علي is not confused with على.

    New in Writing & Content