blader

humanizer

AI#LLM#Natural Language Processing#Writing Assistant#Claude Code

// summary

Humanizer is a specialized skill for Claude Code and OpenCode designed to remove common markers of AI-generated text. It identifies and corrects 29 distinct patterns, such as significance inflation and excessive hedging, to produce more natural writing. Users can also provide personal writing samples to calibrate the tool to match their unique voice and rhythm.

// technical analysis

Humanizer is a specialized skill designed for Claude Code and OpenCode that systematically removes common markers of AI-generated text to produce more natural, human-like writing. By leveraging a comprehensive set of 29 detected patterns—ranging from stylistic quirks like em-dash overuse to structural issues like formulaic hedging—the project aims to strip away the statistical predictability inherent in LLM outputs. A notable technical decision is the inclusion of a two-pass audit process and a voice calibration feature, which allows the tool to adapt its output to match a user's specific writing rhythm and vocabulary.

// key highlights

Detects and removes 29 distinct AI-writing patterns, including promotional language, synonym cycling, and excessive hedging.

Implements a dual-pass rewrite process that includes a final audit to catch lingering AI-isms.

Supports voice calibration by analyzing user-provided writing samples to mimic personal sentence rhythm and word choice.

Integrates directly into Claude Code and OpenCode environments for seamless command-line usage.

Based on empirical data from Wikipedia's 'Signs of AI writing' guide, ensuring the cleanup logic is grounded in observed AI behavior.

Provides clear before-and-after examples for every detected pattern to demonstrate the effectiveness of the transformation.

// use cases

Removing AI-generated stylistic artifacts and filler phrases from technical documentation

Calibrating AI output to match a user's specific personal writing style and tone

Performing an automated audit to detect and rewrite lingering AI-isms in draft content

// getting started

To begin using Humanizer, clone the repository into your ~/.claude/skills/ or ~/.config/opencode/skills/ directory. Once installed, you can trigger the tool using the /humanizer command followed by your text, or by asking the model to humanize specific content directly. For a more personalized result, you can provide a sample of your own writing to calibrate the tool to your unique style.