Humanize – two LLM-agnostic skills to rewrite and detect AI text (github.com)

A new repository has been launched featuring two LLM-agnostic skills, "humanize" and "ai-check," designed to enhance the human-like quality of AI-generated text and detect its origins, respectively. Grounded in over 50 peer-reviewed sources, these skills can be integrated into various LLM agents such as Claude Code and ChatGPT, employing a straightforward installation process. The "humanize" skill applies nine specific techniques from detection literature to transform AI text into a more human-sounding narrative, while "ai-check" provides a forensic examination of text, scoring it across nine signal categories to ascertain its AI-generated likelihood. This development is particularly significant for the AI/ML community as it addresses the increasing challenges in distinguishing AI-generated content from human writing—a concern highlighted by studies revealing automated detectors struggle against paraphrased text. The skills leverage established detection methodologies, including perplexity and burstiness measures, to evaluate and improve writing. However, they also acknowledge limitations; for example, while these skills perform well against certain detectors, they may not evade commercial classifiers that rely on deeper model insights. This dual approach underscores a critical evolution in AI text interaction, balancing the need for authenticity in writing with the growing capabilities of AI technologies.