A Claude Code skill that strips AI writing patterns from text. Detects 30 patterns across five categories, asks for context before rewriting, and runs a self-audit loop to catch anything it missed.
Forked from blader/humanizer -- extended with mode-aware rewriting, 6 additional patterns, and a self-audit loop.
- Mode selection -- asks for register (editorial / professional / technical) and audience before rewriting. A README gets different treatment than a blog post.
- 6 new patterns -- transition word stacking, abstract noun preference, rhetorical questions as transitions, symmetrical paragraph structure, meta-commentary, passive voice overuse.
- 30 patterns total -- up from 24 in v2.x.
- Self-audit loop -- draft, "what's still AI?", final pass.
Clone into your Claude Code skills directory:
mkdir -p ~/.claude/skills git clone https://github.com/Jimmycarroll2021/humaniser.git ~/.claude/skills/humanizerOr copy just the skill file:
mkdir -p ~/.claude/skills/humanizer cp SKILL.md ~/.claude/skills/humanizer/In Claude Code:
/humanizer Paste your text. Claude asks for register and audience, then rewrites.
| Mode | Use for | Approach |
|---|---|---|
| Editorial | Blog, newsletter, essay, social | Personality, first person, rhythm variation |
| Professional | Business doc, proposal, report, email | Fix structure, keep formal register |
| Technical | Docs, spec, README, code comments | Clarity only, no flair |
- Asks for register, audience, and voice preference
- Scans for all 30 patterns
- Rewrites respecting mode rules
- Presents a draft
- Self-audits: "what still reads as AI?"
- Revises and presents the final version
| # | Pattern | Fix |
|---|---|---|
| 1 | Significance inflation | "marking a pivotal moment" -> cut |
| 2 | Notability name-dropping | list of outlets -> specific quote with context |
| 3 | Superficial -ing analyses | "symbolizing... reflecting..." -> remove or source |
| 4 | Promotional language | "nestled within the breathtaking region" -> plain description |
| 5 | Vague attributions | "Experts believe" -> named source with date |
| 6 | Formulaic challenges sections | generic struggles -> specific barrier with data |
| # | Pattern | Fix |
|---|---|---|
| 7 | AI vocabulary | "Additionally... testament... landscape" -> plain words |
| 8 | Copula avoidance | "serves as... functions as" -> "is... has" |
| 9 | Negative parallelisms | "It's not just X, it's Y" -> state the point directly |
| 10 | Rule of three | forced triplets -> natural count |
| 11 | Synonym cycling | "protagonist/character/figure/hero" -> pick one |
| 12 | False ranges | "from X to Y, from A to B" -> list directly |
| 13 | Transition word stacking | "Furthermore, Moreover, Additionally" -> restructure v3 |
| 14 | Abstract noun preference | "utilization" -> "use" v3 |
| 15 | Rhetorical questions as transitions | "But what does this mean?" -> just say it v3 |
| # | Pattern | Fix |
|---|---|---|
| 16 | Em dash overuse | multiple per sentence -> commas or periods |
| 17 | Boldface overuse | OKRs, KPIs -> OKRs, KPIs |
| 18 | Inline-header lists | Performance: Performance improved -> prose |
| 19 | Title case headings | Title Case Every Word -> Sentence case |
| 20 | Emojis in formal writing | remove entirely |
| 21 | Curly quotes | typographic quotes -> straight quotes |
| 22 | Symmetrical paragraph structure | every para same shape -> vary deliberately v3 |
| # | Pattern | Fix |
|---|---|---|
| 23 | Chatbot artifacts | "Great question! I hope this helps!" -> delete |
| 24 | Knowledge-cutoff disclaimers | "details are limited..." -> find source or remove |
| 25 | Sycophantic openers | "Great question! You're right!" -> respond directly |
| 26 | Meta-commentary | "It is worth noting that..." -> just say it v3 |
| # | Pattern | Fix |
|---|---|---|
| 27 | Filler phrases | "In order to", "Due to the fact that" -> "To", "Because" |
| 28 | Excessive hedging | "could potentially possibly" -> "may" |
| 29 | Generic positive conclusions | "The future looks bright" -> specific next step |
| 30 | Passive voice overuse | "was compiled and reviewed" -> name the actor v3 |
- 3.0.0 -- Mode-aware rewriting, audience-first context, self-audit loop, 6 new patterns (30 total)
- 2.2.0 -- Added self-audit loop: draft -> "what's still AI?" -> final pass
- 2.1.1 -- Fixed pattern #18 curly quotes example
- 2.1.0 -- Added before/after examples for all 24 patterns
- 2.0.0 -- Complete rewrite based on Wikipedia content
- 1.0.0 -- Initial release
MIT