Removes AI writing patterns from English text across typography, vocabulary, and structure.
npx skills add https://github.com/asavvin-pixel/unslop --skill unslop-asavvin-pixel
Please help me install this Agent Skill. Skill: unslop Source: https://github.com/asavvin-pixel/unslop Command: npx skills add https://github.com/asavvin-pixel/unslop --skill unslop-asavvin-pixel
references/style-profile.md, so edits match your voice instead of a generic default. - Three edit modes: Free edit (deletes filler, text may shrink), careful edit (preserves structure and 80–110% of length), and minimal edit (only unambiguous AI constructions). - Use Case: Paste an AI-drafted blog post and ask to humanize it; the skill rewrites it with varied sentence rhythm, real specifics only, no self-summarizing paragraphs, and runs a grep-based final check for residual markers. ## Quick Start Ask the AI to unslop or humanize your English text, for example by saying "unslop this" followed by the text you want rewritten.High-intent search queries and answers about installing and using this skill.
Paste the text and ask to humanize or unslop it. The skill rewrites across three levels: typography fixes, a 19-category vocabulary stop list, and structural changes like varied paragraph shapes and honest hedging, then runs a final grep-based check.
A 2026 UMD and Google DeepMind study of 61,608 texts showed surface editing barely reduces structural detection, from 95.5% to 93.9%. The durable signal is how text is built: template paragraphs, chewed conclusions, and invented specifics, which word swaps never touch.
Yes. Say "calibrate to my style" and provide 3–5 texts you wrote yourself, totaling 2,000+ words. The skill extracts your typography, vocabulary, rhythm, and habits into an editable style-profile.md file that overrides defaults on future edits.
No, and that is deliberate. The skill does not target detectors, which err in both directions. Its goal is text that reads well to a human reader, with honest claims and no invented facts, not text engineered to fool a classifier.
The skill cannot add substance that is missing: if the source has no facts, the output is an honest short text plus a list of gaps. It also cannot imitate another named author, and its epistemics rules against invented specifics cannot be switched off.
No, the wordlists and conventions are English-specific. For Russian text, the documentation points to a sister skill called ochelovech, since Russian AI writing has its own lexicon and tells that do not translate.
CloudChefDetect and rewrite AI-generated writing patterns to produce natural human-like text.
hivgb1-aiShips reviewed deliverables through planned distribution channels with human confirmation gates.
lawve-aiVerify factual claims and citations in legal documents against primary sources.
NeoLabHQApplies Strunk's Elements of Style rules to improve documentation clarity and conciseness.