unslop
Also known as: de-AI, humanize AI text, AI text humanizer skill
AI-generated text has recognizable tells: words like 'delve', 'tapestry', 'pivotal', and 'robust' used together, sentences that start with 'Additionally', conclusions that restate everything that just happened. unslop is a skill you install into a coding agent that catches and removes these patterns before they ship. It works on three levels: typography and mechanics, vocabulary and rhetoric, and structure and epistemics.
The project was built by designer Damir Sokolovsky and added to the agent skills ecosystem in mid-2026. It gained significant GitHub traction in late October 2026 as builders started applying it not just to written copy but also to vibecoded UI: the same quality-floor idea that applies to prose applies to layouts that look suspiciously generated. A companion Claude skill called unslop-ui audits screens for generic AI design patterns rather than generic AI prose.
The practical use case is any output that goes to a human audience: marketing copy, product documentation, error messages, landing pages. unslop doesn't just flag problems; it rewrites them. The goal, as the project readme puts it, is text that reads like a thinking person wrote it. It pairs naturally with impeccable, which handles the visual design side of the same anti-slop problem.