A Claude Code skill for removing AI-generated writing traces from scientific text while preserving meaning, accuracy, and scientific rigor.
This project provides a skill (humanAIzer) that rewrites AI-generated text to sound more natural and human-written. It is based on Wikipedia's Signs of AI writing guide.
- AI Pattern Detection: Identifies 24 common AI writing patterns
- Scientific Rigor Preservation: Keeps technical accuracy while humanizing
- ZeroGPT Verification: Uses detect.py to measure AI detection percentage before and after editing
- Quality Scoring: Built-in rubric for evaluating humanized output
- Exaggerated symbolism and promotional language
- Shallow -ing analysis
- Vague attribution
- Overuse of dashes and rule of three
- AI vocabulary (furthermore, key, landscape, etc.)
- Negative parallelism (not just... but also...)
- Excessive connective phrases
- Formulaic section structures
Apply the humanAIzer skill when you need to edit text to remove AI writing traces. The skill will:
- Identify AI patterns in the text
- Rewrite problematic segments
- Preserve the original meaning
- Maintain scientific tone (with slight journalist flair acceptable)
- Measure results with ZeroGPT detection
.claude/skills/humanaizer/
└── SKILL.md # Main skill file with full instructions
humanaizer/scripts/
├── detect.py # ZeroGPT API integration for AI detection
└── suggest_synonims.py # Synonym suggestions (if available)
| Range | Meaning |
|---|---|
| 0–20% | Low AI detection — text sounds human |
| 20–50% | Medium detection — some AI patterns may remain |
| 50–100% | High detection — significant rewriting needed |
Based on humanizer-zh — a guide for removing AI writing traces from text, itself based on Wikipedia's Signs of AI writing guide by WikiProject AI Cleanup.