Skip to content
Back to search

text-humanizer

MarketplaceWorks with Claude CodeGitHub

Atypical Consulting skills for writing that reads like a person wrote it.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/atypical-consulting--texthumanizer/badge.svg)](/listings/atypical-consulting--texthumanizer)

HTML

<a href="/listings/atypical-consulting--texthumanizer"><img src="/api/listings/atypical-consulting--texthumanizer/badge.svg" alt="findsafeskills safety rating" /></a>

Install

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Claude Code

/plugin marketplace add Atypical-Consulting/TextHumanizer
/plugin install text-humanizer@text-humanizer

Then just paste a text and say it sounds like AI, or invoke it directly. The plugin ships one skill per language, and each one only handles texts in its own language:

/text-humanizer:text-humanizer-en   English texts
/text-humanizer:text-humanizer-fr   French texts

Claude.ai and other agents

Download the folder for your language, [`text-humanizer-en`](plugins/text-humanizer/skills/text-humanizer-en/) or [`text-humanizer-fr`](plugins/text-humanizer/skills/text-humanizer-fr/), keep `references/` and `scripts/`, zip it and upload it as a skill in Claude.ai, or place it in your agent's skill directory. Both are self-contained.

Using the scanner on its own

python3 plugins/text-humanizer/skills/text-humanizer-en/scripts/scan.py text.txt --lang en --format linkedin
python3 plugins/text-humanizer/skills/text-humanizer-fr/scripts/scan.py --compare before.txt after.txt --lang fr --format email

Formats: `linkedin`, `email`, `youtube`, `blog`, `description`. Python 3.9+, no dependencies. Each skill folder carries an identical copy of `scan.py`; its output labels are in French.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 18 lines
  1. {
  2. "name": "text-humanizer",
  3. "description": "Atypical Consulting skills for writing that reads like a person wrote it.",
  4. "owner": {
  5. "name": "Atypical Consulting",
  6. "url": "https://github.com/Atypical-Consulting"
  7. },
  8. "plugins": [
  9. {
  10. "name": "text-humanizer",
  11. "source": "./plugins/text-humanizer",
  12. "description": "Rewrites LLM-generated text (FR/EN) so it reads human: removes AI tics, applies format best practices (LinkedIn, email, YouTube script, blog, description), fact-checks dated claims and measures before/after. One skill per language, each handling only its own: text-humanizer-en for English texts, text-humanizer-fr for French texts.",
  13. "version": "1.2.0",
  14. "author": { "name": "Atypical Consulting" },
  15. "license": "MIT"
  16. }
  17. ]
  18. }
README.md · README · 98 lines
  1. # TextHumanizer
  2. **A Claude skill that rewrites LLM-generated text so it reads like a person wrote it. French and English, with one skill per language of your request.**
  3. [![MIT](https://img.shields.io/badge/licence-MIT-blue.svg)](LICENSE)
  4. [![Claude Code plugin](https://img.shields.io/badge/Claude%20Code-plugin-d97757.svg)](https://code.claude.com/docs/en/plugins)
  5. [![Site](https://img.shields.io/badge/site-GitHub%20Pages-222.svg)](https://atypical-consulting.github.io/TextHumanizer/)
  6. Paste a LinkedIn post, an email, a YouTube script, a blog post or a description that ChatGPT, Claude or Gemini wrote. You get back a version that keeps every fact and sounds like you, plus the numbers that show what changed.
  7. It is not a detector-evasion tool. AI tics are symptoms of soft writing: generic openers, predictable structure, borrowed enthusiasm, a flat rhythm. Fix the writing and the tics go with it.
  8. ## What it does
  9. 1. **Measures the original.** `scan.py` counts em dashes, emoji bullets, Markdown residue, typical AI phrases (around 50 per language), sentence-length variation and paragraph uniformity, checks the target format's constraints and gives an indicative score out of 100.
  10. 2. **Checks dated facts.** Software versions, support status, release dates, prices, statistics. It searches primary sources, keeps the author's own choices in the text and flags them with a source and a ready-to-use alternative, and corrects general claims that are simply wrong.
  11. 3. **Rewrites** with the best practices of the format: hook length and hashtags on LinkedIn, 30 to 90 words for a follow-up email, a hook in the first 15 seconds of a video script, 70-word paragraphs and informative headings for a blog post.
  12. 4. **Measures again** and fixes what is left.
  13. 5. **Reports**: the rewritten text, a before/after table, what changed, facts to check, and the concrete detail you could add. It never invents an anecdote, a figure or a quote to "sound human".
  14. ## Results
  15. Four test texts (LinkedIn FR, follow-up email EN, YouTube script FR, blog post EN), 65 checks in total, run with and without the skill on the same model.
  16. | | With skill | Without skill |
  17. |---|---|---|
  18. | Checks passed (final iteration) | **65 / 65** | 49 / 65 |
  19. | Invented details in the rewrite | 0 of 4 texts | 3 of 4 texts |
  20. | Before/after measures in the answer | 4 of 4 | 0 of 4 |
  21. | Indicative score (before → after) | 30-54 → 94-100 | not measured |
  22. Details and method: [BENCHMARK.md](BENCHMARK.md). The test cases live next to each skill, the English ones in [`text-humanizer-en/evals/`](plugins/text-humanizer/skills/text-humanizer-en/evals/) and the French ones in [`text-humanizer-fr/evals/`](plugins/text-humanizer/skills/text-humanizer-fr/evals/), and run with `claude plugin eval`.
  23. ## Install
  24. ### Claude Code
  25. ```bash
  26. /plugin marketplace add Atypical-Consulting/TextHumanizer
  27. ```
  28. ```bash
  29. /plugin install text-humanizer@text-humanizer
  30. ```
  31. Then just paste a text and say it sounds like AI, or invoke it directly. The plugin ships one skill per language, and each one only handles texts in its own language:
  32. ```
  33. /text-humanizer:text-humanizer-en English texts
  34. /text-humanizer:text-humanizer-fr French texts
  35. ```
  36. ### Claude.ai and other agents
  37. Download the folder for your language, [`text-humanizer-en`](plugins/text-humanizer/skills/text-humanizer-en/) or [`text-humanizer-fr`](plugins/text-humanizer/skills/text-humanizer-fr/), keep `references/` and `scripts/`, zip it and upload it as a skill in Claude.ai, or place it in your agent's skill directory. Both are self-contained.
  38. ### Using the scanner on its own
  39. ```bash
  40. python3 plugins/text-humanizer/skills/text-humanizer-en/scripts/scan.py text.txt --lang en --format linkedin
  41. python3 plugins/text-humanizer/skills/text-humanizer-fr/scripts/scan.py --compare before.txt after.txt --lang fr --format email
  42. ```
  43. Formats: `linkedin`, `email`, `youtube`, `blog`, `description`. Python 3.9+, no dependencies. Each skill folder carries an identical copy of `scan.py`; its output labels are in French.
  44. ## What is in here
  45. ```
  46. plugins/text-humanizer/skills/
  47. ├── text-humanizer-en/ English texts only
  48. │ ├── SKILL.md the procedure: measure, fact-check, rewrite, measure, report
  49. │ ├── references/
  50. │ │ ├── markers-en.md AI markers in English, by family, with fixes
  51. │ │ └── formats.md sourced best practices and numeric targets per format
  52. │ ├── scripts/scan.py the measuring tool (report or before/after table)
  53. │ └── evals/ the two English test cases, 16-17 graders each
  54. └── text-humanizer-fr/ French texts only, same procedure written in French
  55. ├── SKILL.md
  56. ├── references/ marqueurs-fr.md, formats.md
  57. ├── scripts/scan.py identical copy
  58. └── evals/ the two French test cases
  59. site/ brief and real examples for the site
  60. docs/ the GitHub Pages site (built output)
  61. ```
  62. ## Honest limitations
  63. - The score is indicative. It counts patterns; it does not judge whether a text is good, and it is not an AI detector.
  64. - Fact-checking needs a web search tool. Without one, dated facts are flagged as unverified rather than checked.
  65. - Best-practice numbers come from vendor studies (Hunter, Saleshandy, vidIQ, Nielsen Norman Group...). They are reference points; an author's proven habit wins.
  66. - Voice matching is only as good as the samples you give it. Paste two or three of your own texts for the best result.
  67. ## En français
  68. Ce skill réécrit un texte généré par une IA pour qu'il sonne écrit par un humain, en français ou en anglais, avec un skill par langue (`text-humanizer-fr` pour le français). Il mesure le texte, vérifie les faits datés, applique les bonnes pratiques du format (post LinkedIn, email, script YouTube, article, description), puis rend le texte réécrit avec un tableau avant/après. Pour une demande rédigée en français, c'est le skill `humaniser-texte` (`/text-humanizer:humaniser-texte`). Installation : voir plus haut.
  69. ## Licence
  70. MIT. See [LICENSE](LICENSE).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.