Skip to content

feat: add humanizer-pl skill (Polish signs of AI writing + humanizer) - #46

Open
odwrotnie wants to merge 1 commit into
softaworks:mainfrom
odwrotnie:feat/humanizer-pl
Open

feat: add humanizer-pl skill (Polish signs of AI writing + humanizer)#46
odwrotnie wants to merge 1 commit into
softaworks:mainfrom
odwrotnie:feat/humanizer-pl

Conversation

@odwrotnie

Copy link
Copy Markdown

What

A Polish-language companion to the existing humanizer skill: skills/humanizer-pl/.

Polish AI text has tells that an English-oriented humanizer misses — English punctuation calques (em dashes instead of the Polish en dash, "English" instead of „Polish" quotation marks, commas after sentence-initial adverbials), syntactic calques (zaadresować problem, dedykowany dla, wydaje się być), and a distinct set of overused phrases (w dzisiejszym dynamicznie zmieniającym się świecie, warto zauważyć, że, kluczowy, podsumowując).

Contents

  • signs-of-ai-writing-pl.md — a comprehensive, sourced field guide to AI-writing markers in Polish, structured after Wikipedia's Signs of AI writing (like writing-clearly-and-concisely/signs-of-ai-writing.md in this repo) but researched from Polish-language sources:
    • Polish academic research: ChatGPT error typology in Polish essays (LingVaria 2024), the PolEval 2025 "Śmigiel" machine-generated-text detection shared task, the CEAID multilingual detection benchmark, stylometry studies, the national anti-plagiarism system (JSA) methodology;
    • Polish Wikipedia community practice: fabricated references as the top marker, utm_source=chatgpt.com trails, Markdown artifacts in wikitext, bulk access-dates;
    • ~130 documented Polish phrases from SEO/copywriting phrase scanners and detector docs, each cross-checked across independent sources;
    • Polish typography norms (PWN) vs. LLM behavior: dash, quotes, comma calques, Title Case;
    • per-model signatures (ChatGPT eras, Claude, Gemini, DeepSeek, Polish Bielik/PLLuM) with marker "expiry dates";
    • a machine-translation vs. native-generation distinction (4 text profiles) — protecting human authors whose text went through DeepL from false accusations;
    • an "ineffective indicators" section (detector false-positive rates are >2× higher for Polish than English per vendor data).
  • SKILL.md — editor instructions that strip these artifacts from Polish text and normalize Polish typography, with before/after examples; flags unverifiable claims as [DO WERYFIKACJI] instead of paraphrasing potential hallucinations.
  • README.md, marketplace.json entry (utilities, keywords incl. polish), built dist/ plugin (python3 scripts/build_plugins.py run).

Notes

  • The guide document is in Polish (its audience edits Polish text); SKILL frontmatter description and README are in English per repo conventions.
  • Structure and universal markers adapted from Wikipedia's Signs of AI writing (CC BY-SA 4.0, attributed in the document); approach modeled on the humanizer skill.

🤖 Generated with Claude Code

Polish-language companion to the humanizer skill:

- signs-of-ai-writing-pl.md: a sourced field guide to AI-writing markers
  specific to Polish (counterpart of Wikipedia's Signs of AI writing),
  compiled from Polish academic research (LingVaria error typology,
  PolEval 2025 Śmigiel, CEAID, stylometry), Polish Wikipedia community
  practice, SEO/copywriting phrase scanners, AI-detector docs (JSA,
  plagiat.pl), education and social-media observations. Covers lexical,
  syntactic, typographic (Polish quotes/dashes/comma calques), structural,
  factual and technical markers, per-model signatures, the machine
  translation vs generation distinction, and ineffective indicators.
- SKILL.md: editor instructions to strip these artifacts from Polish text
  and normalize Polish typography, with before/after examples.
- README.md, marketplace.json entry, built dist/ plugin.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant