I wanted to score my AI coding prompts without sending them anywhere — built a local scoring tool using NLP research papers, Ollama optional

Quick context: I use AI coding tools daily - Claude Code, Cursor, Aider, Gemini CLI. After 6 months I had thousands of prompts in session files and wanted to know which ones actually worked well. Every analytics tool I found either required an account or wanted to send my data somewhere. My prompts contain file paths, internal function names, error messages from production systems. That's essentially a map of my codebase. Not sending that to an API to get scored. So I built reprompt. It runs entirely on your machine. Here's the privacy picture: The default backend is TF-IDF (scikit-learn.