What problem does it solve? AI-generated Russian text carries recognizable traces: copy-paste artifacts like citation markers and tracking URLs, invisible Unicode characters, and stylistic patterns such as forced triads, vague praise, and bureaucratic phrasing. This Skill finds those traces and rewrites the text into natural human prose without distorting facts or adding content the author never wrote. ## Core Features & Use Cases - Marker Detection: 40 testable regex markers in two classes catch hard copy-paste artifacts (ChatGPT citation tags, Gemini span markers, Perplexity S3 links, zero-width characters) and contextual indicators requiring human judgment. - Pattern-Based Rewriting: 56 documented patterns of machine writing (25 base plus 31 Russian-specific) guide selective edits or full rewrites from a factual skeleton, with genre modes for essays, news, correspondence, and technical text. - False-Positive Protection: A decision tree and adversarial corpus ensure live human writing is never flagged or altered; soft signals calibrate edit scope but never produce an authorship verdict. - Use Case: Paste a marketing draft that reads like typical LLM output. The Skill strips stock phrases, emoji-list scaffolding, and hidden metadata, returning only the facts that were already present, rewritten in natural Russian. ## Quick Start Ask the agent to humanize a pasted Russian text with the /humanize command, or run /audit to check it for AI traces without editing.
