⏳ This skill is pending AI review.
Scores will appear once the review pipeline completes.
humanizer-it
Skill for humanizing Italian-language text. Removes signs of AI generation and brings the text back to life. Use ALWAYS when the user asks to: humanize Italian text, remove AI/neural-network traces, make text sound natural/alive, rewrite it like a human, 'umanizzare', remove burocratese, remove fluff, make it less formal. Also use when the user pastes Italian text and says things like 'riscrivi', 'rendilo migliore', 'sembra un robot', 'troppo artificiale'. Works ONLY with Italian. For Russian use humanizer-ru; for English use the original humanizer. Do NOT use for: translation, writing from scratch, grammar-only, code.
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
Not yet listed on ClawHub or SkillsMP
// README
humanizer-it: umanizza il testo IA italiano per Claude Code, Cursor, Codex e altri agenti
🇬🇧 English · 🇮🇹 Italiano
Claude Code / Cowork plugin. Kills AI smell in Italian text. The English humanizer won't help here, and neither will humanizer-ru. Italian AI markers are their own beast: the «IA-taliano» that Treccani officially catalogued in 2023 — English-syntax calques (impronte algoritmiche dell'inglese, A.-M. De Cesare), nominal-style burocratese, the missing intercalari (allora, insomma, magari) and ci/ne clitics that make Italian sound alive, and the flattened antifrasi irony.
Quick start, one command for any agent (Claude Code, Cursor, Codex, Cline and dozens more):
npx skills add ilyautov/humanizer-it
Before: «Nel mondo di oggi l'intelligenza artificiale riveste un ruolo sempre più importante. È importante notare che questa tecnologia rappresenta un potente strumento per l'ottimizzazione dei flussi di lavoro.»
After: «Nell'ultimo anno ho messo strumenti IA in tre progetti. Due sono andati il doppio più veloci. Il terzo è saltato, perché il team ha smesso di controllare quello che sputava il modello.»
Three hard bans in two sentences («Nel mondo di oggi», «riveste un ruolo», «È importante notare che»). Full breakdown below.
Italian fork of the battle-tested humanizer-ru. Same language-independent core (perplexity/burstiness, modes, contrastive subtraction, fact-lock, quad-pass audit); the marker content is rebuilt for Italian and grounded in the research in CORPUS-MARKERS-IT / CORPUS-DESIGN-IT.
What you get
A catalog of 52 patterns across 12 categories: burocratese / nominal style (-zione, -mento), English calques and translationese, the copula «è» overuse, pro-drop violations (redundant io/tu/lui) and under-used ci/ne clitics, falsi amici (realizzare, eventualmente, attualmente), emotional sterility, persuasion tricks, information rhythm, plus the 2025-2026 stylistic fingerprints (jagged-meditation single-word sentences, pseudo-Socratic Q&A chains, decorative emoji per list item, pseudo-therapeutic register). 15 hard-banned constructions that scream "an LLM wrote this", led by the negative parallelism «non solo… ma anche» and the em-dash (in native Italian it's rarer than in English — so more suspicious, and detectors count its frequency).
It also does what a detector can't: it brings back living Italian — intercalari, ci/ne clitics in idioms, left dislocation («Il caffè lo prendo dopo»), litote and antifrasi irony, real idioms. That's the differentiator versus a detector-only tool like aipatterndetector.it.
A research-backed section explains how detectors actually work (perplexity, burstiness, native-Italian morphosyntax) with the Italian benchmark line: DeSegMa-IT @ EVALITA 2026 (UmBERTo ~0.9458) and the key insight that native-from-scratch models (Minerva-7B) are caught at only ~50% recall while English-first models leak at >90% — so the humanizer pushes text toward the native-Italian centroid and silences the anglo-calques.
Deterministic scanner included — zero dependencies
The skill ships with scripts/scan.py, the machine half of Audit mode. It counts what an LLM eyeballs: hard bans, marker categories, sentence rhythm (burstiness), and Italian morphosyntax (nominalization density, pro-drop, ci/ne clitics). Unlike the RU version it needs no spaCy or pymorphy — plain Python stdlib, so it runs anywhere:
python3 skills/humanizer-it/scripts/scan.py file.txt
echo "il tuo testo" | python3 skills/humanizer-it/scripts/scan.py -
It prints a CLEANLINESS: N/100 score and a band (clean / edit / rewrite). See eval/ for the harness and eval/RESULTS.md for before/after deltas on a stratified corpus (7/7 AI texts cleaned to zero hard bans; 3/3 human Wikipedia controls pass clean).
Install
1. Claude.ai (Web UI)
- Download the packaged skill: humanizer-it.zip
- Open Claude.ai → Settings → Capabilities → Skills.
- Click Upload skill and select the downloaded archive.
Don't take
archive/refs/heads/main.zip: the uploader wants the skill folder at the archive root, and in the repo archive it sits insidehumanizer-it-main/skills/humanizer-it/. The release ZIP is 92 KB of skill; the repo archive is ten times that, most of it website and eval corpus. To build it yourself:python scripts/build_release_zip.py --output dist/humanizer-it.zip .
2. Organizations (Enterprise & Team)
Workspace admins can roll the skill out via Admin Console → Workspace Skills → Add skill. Upload the same ZIP, no per-user installation needed.
3. Claude Code, Cowork, API (local agents)
Plugin marketplace (recommended):
/plugin marketplace add ilyautov/humanizer-it
/plugin install humanizer-it@humanizer-it
Manual:
git clone --depth 1 https://github.com/ilyautov/humanizer-it /tmp/humanizer-it
mkdir -p ~/.claude/skills
cp -r /tmp/humanizer-it/skills/humanizer-it ~/.claude/skills/
Copy the whole folder, not just SKILL.md: the skill ships with the deterministic scanner (scripts/scan.py). No pip install needed — it's stdlib-only.
4. Codex CLI (OpenAI)
git clone --depth 1 https://github.com/ilyautov/humanizer-it
mkdir -p ~/.codex/skills
cp -r humanizer-it/skills/humanizer-it ~/.codex/skills/
Restart Codex after installing; invoke with $humanizer-it or let it auto-trigger.
5. Other agents (shared SKILL.md standard)
The Agent Skills format is cross-platform. Other agents (Copilot, Cline, Roo Code, Goose, OpenCode, Cursor, Gemini CLI, …) read the same SKILL.md: copy the skills/humanizer-it folder into the agent's skills dir and restart.
Modes
- Full edit (default): all 52 patterns, voice calibration, quad-pass audit.
- Audit: diagnosis only, returns detected patterns with priority A-D and a cleanliness score.
- Targeted fix: works on a specific category only.
Usage
Ask in Italian:
Umanizza questo testo: [paste text]
Riscrivilo, sembra un robot: [paste text]
Triggers: "umanizza", "togli i segni dell'IA", "rendilo naturale", "sembra artificiale", "riscrivi come un umano".
Before / After
Before:
Nel mondo di oggi l'intelligenza artificiale riveste un ruolo sempre più importante. È importante notare che questa tecnologia rappresenta un potente strumento per l'ottimizzazione dei flussi di lavoro.
After:
Nell'ultimo anno ho messo strumenti IA in tre progetti. Due sono andati il doppio più veloci. Il terzo è saltato, perché il team ha smesso di controllare quello che sputava il modello.
Three hard bans in two sentences («Nel mondo di oggi», «riveste un ruolo», «È importante notare che»). Typical.
Do AI detectors work on Italian?
Better than on most languages, actually — Italian has a native detection line (DeSegMa-IT @ EVALITA
// HOW IT'S BUILT
KEY FILES