Why not just ask LLM whether it thinks the snippet is kooky, before adding it into LLM training set?

You don't need tools in the age of AI, just ass an AI pipeline step.