A Claude Code skill that detects and removes AI-generated writing traces from English academic prose. Lightweight, token-efficient, and ready to use out of the box.
A concise set of editing rules that guide an LLM to identify and eliminate AI writing patterns in academic papers. The rules are built around patterns recognized by AIGC detectors: the goal is to change the patterns, not the content.
This project prioritizes brevity, accuracy, and token economy — no lengthy examples or scoring systems, just the rules that work directly. Paste and use; no wasted context.
| File | Format | Description |
|---|---|---|
SKILL.md |
Claude Code Skills | YAML frontmatter, installable to ~/.claude/skills/ |
SYSTEM_PROMPT.md |
System Prompt | Plain prompt text, works with any LLM |
Both files contain the same core content. Pick the format that fits your workflow.
npx skills is an AI agent skill manager supporting Claude Code, Cursor, Codex, and 40+ AI coding assistants. Install with one command:
npx skills add Duducoco/humanizer-academic-enAfter installation, the skill is automatically linked to your AI coding assistant. Verify:
npx skills listUpdate to the latest version:
npx skills update Duducoco/humanizer-academic-enUninstall:
npx skills remove Duducoco/humanizer-academic-en-
Download or clone this repository:
git clone https://github.com/Duducoco/humanizer-academic-en.git
-
Copy
SKILL.mdto Claude Code's skills directory:- macOS / Linux:
~/.claude/skills/ - Windows:
%USERPROFILE%\.claude\skills\
cp humanizer-academic-en/SKILL.md ~/.claude/skills/humanizer-academic-en.md - macOS / Linux:
-
Restart Claude Code or reload skills. Verify by typing:
/humanizer-academic-en
Only the SKILL.md file is needed — no need to clone the full repository:
curl -o ~/.claude/skills/humanizer-academic-en.md \
https://raw.githubusercontent.com/Duducoco/humanizer-academic-en/main/SKILL.mdCopy the entire contents of SYSTEM_PROMPT.md and paste it into the System Prompt field of Claude, ChatGPT, DeepSeek, or any other LLM. The skill activates in any conversation.
No installation required — paste and go.
After activating the skill, paste your paper excerpt directly:
/humanizer-academic-en Please polish the following passage:
[paste your text here]
You can also omit the prefix in skill-activated sessions:
Please polish the following passage:
[paste your text here]
After setting the system prompt, send any text directly — no additional commands needed.
The skill targets 29 distinct AI writing patterns, including:
- Inflated significance language ("pivotal", "underscores the broader significance")
- Promotional adjectives ("groundbreaking", "remarkable", "compelling")
- Formulaic scaffolding ("not only... but also...", rule-of-three compulsion)
- Vague attribution ("studies show", "scholars argue" without citations)
- Fake-depth sentence endings ("thereby advancing the field...")
- Overused AI vocabulary clusters ("moreover", "notably", "showcase")
- Em dash abuse (banned in academic prose)
- Excessive hedging or hedging stacks
- Passive-voice saturation
- Nominalization stacking and heavy noun-based phrasing
- Sentence-length and clause-depth overload
- Synonym cycling to avoid repetition
- Chatbot residue ("I hope this helps", "as of [date]")
- Generic optimistic conclusions
The output preserves meaning, citations, and disciplinary register while restoring natural authorial voice.
- Hard mode (default): shorter sentences, minimal modifiers, strict de-hype
- Soft mode: preserves more stylistic texture while still removing AI traces
- npx skills — package distribution
- Wikipedia: Signs of AI writing (WikiProject AI Cleanup)
- blader/humanizer
- op7418/Humanizer-zh
- redbaronyyyyy-eng/humanizer-zh-academic
MIT License