wilu222

Unpolish AI Writing — AI skill for Claude Code

AI community

Agent skill that removes signs of AI-generated writing, and optionally adds sparse typing errors using probability distributions informed by researched human error patterns.

How to install Unpolish AI Writing

This entry records only its repository, not the path inside it, so there is no exact command to give. Open wilu222/unpolish-ai-writing and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Unpolish AI Writing does

Agent skill that removes signs of AI-generated writing, and optionally adds sparse typing errors using probability distributions informed by researched human error patterns.

Alternatives in AI

Humanizer — A Claude Code skill that removes signs of AI-generated writing from text, making it sound more natural and hum 11.3k ★
Firecrawl MCP Server — 🔥 Official Firecrawl MCP Server - Adds powerful web scraping and search to Cursor, Claude and any other LLM c 6.1k ★
Surface — Show ranked attack surface for a target from its recon manifest + hunt memory 3.8k ★

README

unpolish-ai-writing

Removes signs of AI-generated writing from text using two methods:

Detect and remove common signs of AI writing
Optionally adds sparse typing errors informed by researched human error patterns.

How it works

Step 1 (Default): AI detection & cleanup

Detect common AI writing patterns (view NOTES.md for more info)

**Selected examples**

Pattern	Before	After
Not-X-it's-Y	"It's not just X — it's Y"	State Y directly
Rule of three	"skilled, attentive, and precise"	Keep the real details; drop the slogan beat
Empty significance / magic adverb	"quietly transformative", "truly special"	Cut, or name what actually changed
Invented concept label	"the X paradox/trap/creep" (made-up jargon label)	Plain description, or drop the label
Copula dodge	"stands as a testament to…"	"is…"
Em-dash habit	"worth it—even with the queue—"	Periods, commas, or a shorter sentence
Punchy fragment stack	"Less busywork. More impact. Better focus."	One concrete sentence

Step 2 (Optional): human-like writing errors

Same cleaned input always gets the same typos. “Reroll” changes the seed so you get a different set. About one error per 600 words, and often none.

Convert the input draft (plus a reroll phrase, if asked) into a hash value
Use that number to decide how many imperfections to add (Poisson from word count)

![Poisson distribution of number of errors by word count](./video/poisson.gif)

If the count is above zero, hash again once per imperfection to pick type and placement, using a weighted distribution

![Distribution of imperfection type](./video/tape.gif)

Type	Example
Misspelling / wrong word	`could have` → `could of`, `affect` → `effect`
Dropped apostrophe	`don't` → `dont`, `I'm` → `Im`
Missing end punctuation	last sentence loses its `.` or `?`
Uncapitalized sentence start	`The` → `the`
Extra space	`I think