# What Is Humanizer for AI Agents? A Portable Skill for Natural Text Rewriting > Discover Humanizer for AI agents. This skill rewrites AI text to sound human using 25 stylistic patterns to eliminate AI-telling constructs. Get natural sounding text. - Repository: [Siqi Chen/humanizer](https://github.com/blader/humanizer) - Tags: getting-started - Published: 2026-09-10 --- **Humanizer is a Markdown-based skill that rewrites AI-generated text to sound authentically human by applying 25 stylistic patterns that detect and eliminate common "AI-telling" constructs.** Humanizer is an open-source tool from the `blader/humanizer` repository that functions as a portable skill for any AI agent supporting the Skills CLI format. Unlike traditional software requiring complex installation, Humanizer operates entirely through a single Markdown file containing the prompt logic and metadata that agents consume to transform robotic language into natural prose. ## How Humanizer Detects and Rewrites AI Text Humanizer works by scanning input text against a **catalog of 25 stylistic patterns** grouped into five distinct sections. These patterns identify specific "AI-telling" constructs—such as staged introductions, forced triads, over-used buzzwords, decorative formatting, and residual chatbot phrasing—that signal machine-generated content. The rewriting engine follows a structured process: 1. **Detection**: Scans the input and marks the strongest "tells" first 2. **Drafting**: Creates a rewrite that respects the original facts while removing identified patterns 3. **Validation**: Iteratively checks the draft against the pattern catalog to ensure no AI artifacts remain 4. **Output**: Produces a cleaned version that retains the author's voice when a writing sample is supplied According to the source code in [`README.md`](https://github.com/blader/humanizer/blob/main/README.md) (lines 77-129), each pattern includes before-and-after examples that guide the transformation, ensuring the output preserves meaning while eliminating mechanical phrasing. ## Installing Humanizer for AI Agents Humanizer supports three installation methods documented in [`README.md`](https://github.com/blader/humanizer/blob/main/README.md) (lines 9-27), requiring no traditional software compilation or dependency management. ### Global Installation via Skills CLI Install Humanizer across all your AI agents using the npm-based Skills CLI: ```bash npx skills add blader/humanizer --global ``` ### Claude Plugin Installation For Claude-specific implementations, install through the plugin marketplace: ```text /plugin marketplace add blader/humanizer /plugin install humanizer@humanizer ``` ### Manual Installation Copy the core skill file directly into your agent's skill folder: - Copy [`SKILL.md`](https://github.com/blader/humanizer/blob/main/SKILL.md) from the repository into your agent's designated skills directory - The file contains all necessary YAML metadata (`metadata.version`, `name`, etc.) and functional logic required for immediate execution ## Using the Humanizer Skill Once installed, invoke Humanizer using the `/humanizer` command or natural language requests. The skill accepts raw text, file paths, or voice samples for style matching. ### Basic Text Rewriting Paste AI-generated content directly after the command: ```text /humanizer [Paste your AI‑generated text here] ``` Or use plain language: ```text Please humanize this text: [your text] ``` ### Voice Matching with Writing Samples To maintain consistency with your personal writing style, provide a sample before processing: ```text /humanizer Here's a sample of my writing for voice matching: [2‑3 paragraphs of your own writing] Now humanize this text: [AI‑generated text you want to clean up] ``` ### In-Place File Editing Process documents without leaving your workspace: ```text Humanize the prose in docs/launch-post.md ``` ### Three-Step Output Format Every invocation returns a structured response: - **Initial rewrite**: The first pass at removing AI patterns - **Critique**: A brief analysis of any remaining AI-like phrasing - **Final version**: The polished text ready for use ## Core Architecture and Source Files Humanizer's architecture consists entirely of Markdown documents, making it uniquely portable across AI agent platforms. ### SKILL.md as the Runtime Engine The file [`SKILL.md`](https://github.com/blader/humanizer/blob/main/SKILL.md) serves as both documentation and executable code. It contains: - **YAML frontmatter**: Metadata including `metadata.version` and `name` that the Skills loader consumes - **Plain-language prompts**: The complete functional logic that drives the 25-pattern detection system As implemented in `blader/humanizer`, this single-file architecture means agents load the skill by parsing Markdown rather than importing libraries or executing binaries. ### README.md Pattern Library The [`README.md`](https://github.com/blader/humanizer/blob/main/README.md) file (lines 66-73) embeds the complete **25-pattern catalog** that forms the decision-making engine. Each pattern describes specific "tells" such as "Not X but Y" constructions or forced triads, accompanied by transformation examples that the skill references during rewriting. ### AGENTS.md Compatibility Layer The [`AGENTS.md`](https://github.com/blader/humanizer/blob/main/AGENTS.md) file provides guidelines for maintaining compatibility across different agent platforms and instructions for updating the skill without breaking its package format, ensuring consistent behavior whether running in Claude, OpenAI agents, or custom CLI implementations. ## Summary - Humanizer for AI agents operates as a **Markdown-based skill** requiring no traditional code deployment or API keys - The tool applies **25 stylistic patterns** across five categories to identify and eliminate AI-generated phrasing - Installation options include **global CLI deployment**, **Claude plugins**, or **manual file copying** of [`SKILL.md`](https://github.com/blader/humanizer/blob/main/SKILL.md) - Users receive **three-stage output**: an initial rewrite, a critique of remaining tells, and a final polished version - Voice matching capabilities allow the skill to **adapt to personal writing styles** when provided with sample text - All functional logic resides in portable Markdown files, enabling **cross-platform compatibility** without code changes ## Frequently Asked Questions ### What makes Humanizer different from other AI text humanizers? Unlike SaaS tools or Python libraries requiring API calls and dependencies, Humanizer for AI agents is a **skill-not-software** solution. It consists entirely of a [`SKILL.md`](https://github.com/blader/humanizer/blob/main/SKILL.md) file containing prompts and metadata that any Skills CLI-compatible agent can load and execute without additional infrastructure or API costs. ### Can Humanizer match my specific writing style? Yes. When you provide **2-3 paragraphs of your own writing** before the text to be humanized, the skill performs voice matching. It analyzes your sentence structure, vocabulary choices, and stylistic tendencies, then applies those characteristics to the rewrite while still eliminating AI-specific patterns. ### What specific phrases does Humanizer target? The 25 patterns target concrete "AI tells" including **staged introductions** ("In the world of..."), **forced triads** (grouping concepts into threes unnaturally), **decorative formatting** (overuse of bold and bullet points), **residual chatbot phrasing** ("As an AI language model..."), and specific syntactic constructs like **"Not X but Y"** formulations. ### Do I need programming skills to install Humanizer? No programming knowledge is required. You can install Humanizer using simple terminal commands like `npx skills add` or by manually copying the [`SKILL.md`](https://github.com/blader/humanizer/blob/main/SKILL.md) file into your agent's skills folder. The entire tool operates through conversational commands such as `/humanizer` or natural language requests like "Please humanize this text."