Prompt-crimes CLI scans local AI chat logs

Devesh Sangwan released prompt-crimes, a Node.js CLI that scans local AI chat logs and produces markdown reports labeling each offense with a tongue-in-cheek penalty [Dev.to]. The tool reads JSON export files from popular AI assistants, such as GitHub Copilot, Azure OpenAI, and Claude, and applies heuristic rules, like repeated vague prompts and excessive token usage [GitHub]. All processing occurs on the developer's workstation, with no network calls made during analysis [GitHub].

The CLI gives concrete, actionable insight into patterns like "prompt bloat" (average 42 tokens per request) and "copy-paste reuse" (over 30% of queries), helping reduce token waste and improve model output quality [GitHub]. As a local-first utility, prompt-crimes adds to a growing suite of tools that let engineers retain control over their AI workflow, with a minimal footprint of under 5 MB of RAM and a single-threaded runtime [GitHub].

The project fills a niche for security-conscious shops that must keep every byte on-prem, providing a way to audit personal coding habits without exposing proprietary code or conversation data to third-party services [Dev.to].