A Claude skill that scrubs 30 patterns of AI-speak

What it does This is essentially a SKILL.md file that teaches Claude Code and OpenCode to rewrite AI-generated text by removing its mechanical fingerprints. It encodes 30 specific anti-patterns drawn from Wikipedia’s “Signs of AI writing” guide—things like significance inflation, sycophantic chatbot sign-offs, and the dreaded rule of three. It can also calibrate to your own writing samples so the output mimics your sentence rhythm and quirks instead of producing generic “clean” text.

The interesting bit The project treats AI prose as a predictable dialect with identifiable tics rather than a vague vibe problem. Each pattern includes a concrete before-and-after example—“serves as an enduring testament” becomes “is”—and the skill runs a second audit pass to catch any lingering AI-isms the first rewrite missed.

Key highlights

• Detects 30 patterns across content, language, style, communication, and filler/hedging categories
• Based on Wikipedia’s WikiProject AI Cleanup field guide, which catalogs thousands of observed AI text instances
• Includes voice calibration: feed it 2–3 paragraphs of your own writing and it will match your personal style
• Performs a final “obviously AI generated” audit followed by a second rewrite to catch stragglers
• The worked example transforms a wall of buzzwords into plainspoken, specific prose about coding assistants

Verdict Anyone publishing LLM-drafted text they didn’t have time to manually rewrite will find this useful. If you never generate prose with AI, there is nothing here for you to fix.