A systematic audit for AI-generated writing tics

What it does

It’s a portable SKILL.md (and optional detector engine) that drops into Claude Code, Cursor, OpenClaw, Hermes, and other agent tools. It audits text for AI-isms—chatbot openers, significance inflation, copula avoidance, and the like—then either flags them or rewrites the prose to strip them out. You can also set a voice profile to shape whether the result sounds casual, blunt, technical, or warm.

The interesting bit

The rigor is the point. Instead of a vague “make this human” prompt, it runs a two-pass audit against 109 specific word replacements across three severity tiers and 47 pattern categories, then re-reads its own rewrite to catch stragglers. A separate detector layer adds stylometric signals—punctuation distribution, function-word entropy—to catch tells that rule lists miss.

Key highlights

• Three modes: full rewrite, detect-only with P0/P1/P2 severity ratings, and targeted in-place edit.
• Tiered vocabulary table: 109 entries plus 10 multi-word Tier 3 phrases; Tier 1 always flags, Tier 2 flags on clustering, Tier 3 on density or repetition.
• 47 pattern categories covering content inflation, language tics, structural tells, and AI-tool fingerprints like placeholder markup.
• Optional voice profiles (casual, professional, technical, warm, blunt) that apply independently of audience context.
• One SKILL.md runs across Claude Code, Cowork, OpenClaw, Cursor, Hermes, Codex, Windsurf, Cline, Copilot, and ChatGPT Custom GPTs.

Verdict

Worth a look if you publish AI-assisted drafts and want a repeatable, structured way to scrub boilerplate tics. Not useful if you don’t use agent-based writing tools or if you enjoy reading paragraphs that start with “Certainly!”