Make AI-generated content undetectable. Prevents the patterns AI detectors and experienced developers flag - em dashes, buzzwords, over-formatting, narrating comments, git co-author trailers, and uniform sentence structure.
AI detectors look for statistical patterns, not meaning. This plugin prevents those patterns at the source. The agent writes human-sounding content from the start instead of you fixing it after.
npx skills add ofershap/ai-humanizerOr copy skills/ into your .cursor/skills/ or .claude/skills/ directory.
| Type | Name | Description |
|---|---|---|
| Skill | ai-humanizer |
10 rules covering vocabulary, formatting, structure, comments, commits, and git metadata |
| Rule | best-practices |
Always-on behavioral rule that catches AI patterns before they're written |
| Command | /audit |
Scan your project for AI fingerprints in READMEs, docs, comments, and git history |
| AI pattern | Why it gets flagged | What the plugin enforces |
|---|---|---|
| Em dashes (---) everywhere | AI uses them 5-10x more than humans | Hyphens with spaces, periods, or restructured sentences |
| "delve", "leverage", "robust" | 50-700x more frequent in AI text | Plain words: "use", "important", "careful" |
| "In today's...", "Happy coding!" | Opening/closing cliches | Direct, varied openings |
| 12-18 word sentences consistently | Uniform length is a strong signal | Mixed short and long sentences |
// Import the module |
Narrating comments are an instant tell | Only non-obvious intent in comments |
Co-authored-by: Cursor |
Permanently marks your git history | No AI attribution in commits |
| Bold labels on every bullet | Over-formatting signals AI | Sparse, varied formatting |
- think-first - Plan-first workflow (pairs well with humanized output)
If this helped your workflow, a star helps others find it.
MIT