Humanizer is a Markdown-based skill designed to rewrite AI-generated text into a more natural, human-like prose style without altering the original factual claims. Developed by blader, the tool functions by identifying and correcting "AI tells"—statistical patterns common in large language models—such as forced triads, "not X but Y" structures, one-line dramatic closers, and excessive use of qualifiers or sales language. The system utilizes a multi-step workflow: it identifies patterns, drafts a rewrite that breaks the original structure, verifies the draft against the source to prevent hallucination, and produces a final version. To ensure factual integrity, the tool is strictly prohibited from inventing details; if information is missing, it prompts the user instead of guessing. Users can provide writing samples to match a specific voice, rhythm, and punctuation style. Technical implementation is highly flexible, supporting installation via the Skills CLI (`npx skills add blader/humanizer`) for various agents or as a plugin for Claude Code. The tool is designed to preserve code, data, and frontmatter, focusing solely on prose refinement. As of version 3.0.0, the system has consolidated its logic into 25 distinct patterns, prioritizing edits based on the strength and frequency of the detected linguistic patterns.
Read original
github-trending/python