A new open-source CLI tool called Smellcheck takes a refreshingly low-tech approach to detecting AI-generated text. Instead of running content through a machine learning classifier or sending it to a detection API, it scans for four categories of telltale patterns that LLMs leave behind like fingerprints.
The tool checks for typography habits (em dashes, curly quotes, and the Unicode ellipsis character that humans almost never type but LLMs produce constantly), suspicious Unicode like decorative emoji in plain text, buzzword density (phrases like "delve into," "tapestry," and "it's worth noting" that are statistically overrepresented in AI output), and unnaturally formal vocabulary - words like "aforementioned," "heretofore," and "whilst" that people recognize but rarely choose on their own.
It's a Node.js package written in TypeScript, installable via npm install github:fbuchinger/smellcheck. You can run it as a CLI tool against any text file or pipe content through stdin. It also works as an importable library with both sync and async initialization. Exit codes (0 for clean, 1 for flagged) make it easy to wire into git hooks or CI pipelines - useful for teams that want to flag AI-generated pull request descriptions or documentation before they ship.
The important caveat: Smellcheck flags text that looks suspicious, not text that definitively came from an AI. A human writer who loves em dashes will trigger false positives. An AI response carefully prompted to avoid these patterns will slip through. Think of it as a spell-checker for AI tells, not a forensic tool. For teams that just want a fast, private, zero-cost first pass before deciding whether content needs human review, it fills a gap that paid detection APIs don't. MIT licensed.