gitmyhub

code-humanizer

★ 36 updated 5d ago

humanizer, but for code — an agent skill that removes AI-generated code slop: duplicated helpers, try-import fallbacks, broad excepts, speculative abstractions. Test-gated, behavior-preserving.

An AI coding assistant skill that finds and safely cleans up messy patterns left behind by AI-generated code.

MarkdownClaude CodeAgent Skillssetup: easycomplexity 1/5

This project is a skill for AI coding assistants like Claude Code, Cursor, and similar tools. A skill in this context is a set of instructions that teaches the assistant a specific way of working. This particular skill teaches the assistant to find and clean up messy patterns that AI coding tools tend to leave behind in a codebase, things like duplicated helper functions, overly cautious error handling that hides real problems, and abstractions built for only one use case.

The project catalogs sixteen specific patterns split into tiers: copied or reinvented code, unnecessary layers of abstraction, defensive code that swallows errors silently, clutter like leftover comments and unused imports, and weak automated tests. Each pattern comes with an example of what it looks like and how severe it is, and some patterns are marked as acceptable in certain situations rather than automatically flagged as bad.

A key rule the skill follows is that it should never change how the code actually behaves while cleaning it up, including things like which error type gets raised. It only makes edits when a test suite exists to confirm nothing broke, and it makes one type of fix at a time so each change can be reviewed on its own.

Installing it is a single command that adds the skill to whichever coding assistant someone is already using, or it can be copied in by hand since it is just one text file with instructions, no extra software required.

This tool is aimed at developers and teams who use AI coding assistants regularly and want a way to catch and fix the sloppy patterns those assistants tend to introduce, rather than at people who do not write code at all.

Where it fits