AI Tells Fixer — Sound Like You, Not a Chatbot

AI Tells Fixer — Sound Like You, Not a Chatbot

Finds AI tells in English drafts line by line (delve/tapestry, em-dash density, rule of three, "It's not just X", boilerplate openers and closers) and rewrites them with purpose presets for LinkedIn, email, cover letter, blog and product copy. Can learn a voice profile from your own samples. A value-based fact lock proves units, order, dates, times, numbers and names survived. Downloadable skill; runs locally on Python stdlib. No detector-bypass promise.
#الكتابة#التسويق#الإنتاجية
التقييم
مطلوب المزيد من التقييمات
المبيعات
0
كيفية الاستخدام
تنزيل

Sound like you, not ChatGPT - with proof your numbers didn't change.

You can spot an AI draft in one line: "I'm thrilled to announce...", "In today's fast-paced world", "It's not just X, it's Y", Additionally / Furthermore / Moreover on every paragraph, an em dash in every sentence, and "I hope this helps! Feel free to reach out." Your readers spot it too.

Most "humanizers" are a hosted box that returns a different draft and a vague score. This skill does it in the open, on your machine: it counts each habit with line numbers, rewrites with a rulebook for your kind of text, and proves two things with a bundled checker - the tells went down, and every number, price, date, time, name, link and quote survived.

📦 This is a downloadable skill, not a hosted agent. You get the files and run them in your own Claude Code, Codex, OpenClaw or Hermes. Your text never leaves your machine through this skill. The checker needs Python 3.9+ - nothing else, no API key.


What it fixes

  • Stock vocabulary - delve, tapestry, testament, realm, landscape, navigate, leverage, unlock, elevate, seamless, robust, pivotal, crucial, game-changer, myriad, plethora, boasts, showcases, "it's worth noting" - each replaced with the fact it stood for, or cut
  • Stock constructions - "It's not just X, it's Y", "not only... but also", "Whether you're X or Y", rhetorical-question openers, "Here's the thing", "Let's dive in", "In conclusion"
  • Em-dash and rule-of-three habits - counted per 1,000 words, with the line each one is on
  • Connective openers and over-hedging - Additionally / Furthermore / Moreover; may / might / could potentially
  • Boilerplate intros and outros - "I hope this email finds you well", "Feel free to reach out", "Drop a comment below!", cover-letter cliches
  • Stiff, even, over-formatted text - missing contractions in casual copy, sentences all the same length, bold, emoji, bullets, Title Case Headings
  • US/UK spelling mixed in one piece

Five purpose presets change the rules: LinkedIn · email · cover letter · blog · product page. A cover letter keeps its formal "I am"; a product page keeps its bullets.


Your voice, not a generic one

Paste 1-3 things you wrote. The checker builds a voice profile - average sentence length and spread, share of short and long sentences, contraction rate, first- and second-person rate, paragraph length, em-dash / exclamation / question habits - and the rewrite aims at those numbers. scan --voice then reports how far any text is from yours.

It matches habits, never content: nothing is copied from your samples, and no "I tried this for a month" gets invented.


What comes back

  1. The rewrite - clean, ready to paste
  2. A change log - Original | Change | Why, grouped by kind
  3. Before → after measurement - tell index and the families that fell
  4. Fact lock - PASS, or exactly which number / date / name / quote moved
  5. Voice distance (if you gave samples)
  6. Habits to watch - what your draft or your AI tool keeps doing, so your next prompt is better

Sample (from the bundled worked example)

Input: a 206-word LinkedIn launch post.

I'm thrilled to announce that Acme Robotics has officially launched Harbor... In today's fast-paced world, logistics teams are navigating an ever-evolving landscape - and the stakes have never been higher. Harbor isn't just a scheduling tool, it's a game-changer that unlocks seamless collaboration... Moreover, we will host a live demo on November 20, 2025 at 3 PM EST... In conclusion... Drop a comment below!

Scan: tell index 79 / 100 (High) · 15 stock-vocabulary hits · 6 stock constructions · 3 connective openers · 5 boilerplate lines · 2 em dashes

Output:

We launched Harbor, Acme Robotics' warehouse scheduling platform.

It took 14 months and a team of 23 engineers. The result: dock wait times at the pilot sites in Rotterdam and Ohio fell by 37%, and customers saved $2.9 million in detention fees.

Measured: tell index 79 → 2 · vocabulary 15 → 0 · constructions 6 → 0 · boilerplate 5 → 0 · fact lock PASS (14 distinct facts: 14 months, 23, 37% down, $2.9M, 500, Nov 20 2025, 3 PM EST, Acme, Harbor, Rotterdam, Ohio, SAP, Oracle ...) · 37 percent → 37%, $2.9M → $2.9 million counted as the same value · words 206 → 93

With a voice profile built from a 327-word sample of the author's own writing: distance from that voice 73 (AI draft) → 44 (standard rewrite) → 30 (voice-steered rewrite).


The fact lock compares values, not spelling

Written as Counted as
$29,000 · USD 29,000 · twenty-nine thousand dollars the same amount
$2.9M · $2.9 million the same amount
37% · 37 percent · thirty-seven percent the same value
3 PM · 3:00pm · 15:00 · 9 in the morning = 9 AM the same time (3 AM, or 3 PM PST instead of EST, is different)
52 minutes · 52 min · $5 per month · $5/mo the same quantity (52 hours, 5 kg vs 5 lb, $5 per year, up vs down 5% are different)
from $10 to $20 an ORDERED pair: "from $20 to $10" is a CHECK (so are 11/03 vs 03/11, Tuesday vs Thursday, first of May vs third of May)
November 20, 2025 · Nov 20, 2025 · 11/20/2025 · 20 November 2025 the same date
Acme → Apex · $29,000 → $92,000 · 3 PM → 3 AM CHECK - flagged with the exact value

So a rewrite that quietly changes a price, a time, a date or a company name does not get a pass. 23 regression cases ship with the build and all pass.


Honest about what it is

  • Not an AI-detector bypass. It removes the habits that make English read machine-made. It never claims a text is "human", never quotes a detector score, and makes no promise about GPTZero, Turnitin, Originality or any other tool - they disagree with each other and change without notice.
  • No invented content. No new numbers, sources, quotes, stories, keywords or experiences. If a sentence was pure filler it is cut, and the change log says so.
  • The checker counts style; you stay the editor. PASS covers numbers, amounts, dates, times, proper names, links and quotations. It does not check claims or facts stated only in ordinary words, and a name that appears only as the first word of a sentence may be missed. Read the rewrite before you send it.

What you get

english-ai-tells-fixer/
  SKILL.md                         the workflow your agent follows
  README.md                        install + checker reference
  LICENSE.txt
  scripts/en_tells.py              scan · compare · facts · voice (Python stdlib only)
  references/rewrite-rules.md      every flag: why it reads machine-made, the fix, before → after
  references/purpose-presets.md    linkedin · email · cover-letter · blog · product
  references/worked-example.md     a full run, end to end, with and without a voice profile
  references/samples/              the example's drafts, rewrites, reports and voice profile

FAQ

Which texts? LinkedIn posts, work emails, cover letters, blog posts, product pages, newsletters - any English professional prose.

Do I need a voice sample? No. Without one you get the standard rewrite. With one (300+ words of your own writing is best) the rewrite targets your sentence length, contraction habit and rhythm.

Will it pass [detector]? No promise, for any tool. It makes the text read naturally to a human reader; that is what it is built and measured for.

Will it change my numbers? It is told not to, and the fact lock catches it when it does: any changed amount, date, time, name, link or quote shows up as CHECK.

Does it translate or fact-check? No. English in, English out; a wrong claim in the draft stays as written.

Coursework? It will polish your own writing. It will not rewrite someone else's work so it can be submitted as yours.


Limits - please read before buying

  • English only; not a translator, not a fact-checker.
  • Style rewrite, not authorship detection; no detector guarantees.
  • Fact lock checks numbers, amounts, dates, times, proper names, links and quotations - not every claim.
  • Requires an agent client that loads SKILL.md skills; the checker needs Python 3.9+.
  • Licence. Personal or single-organisation use, including commercial work. Redistribution or resale of the files is not permitted. Full terms in LICENSE.txt.