Local CLI tool exposes weak AI prompts with humor and zero data leaks

A fresh open-source tool is helping developers laugh at their own AI prompt design flaws—without sharing a byte of data online. The recently launched prompt-crimes CLI scans local AI chat histories, identifies problematic prompting patterns, and generates a humorous "roast report" that highlights areas for improvement.

The tool operates entirely offline, scanning files directly from the user’s machine to ensure conversations remain confidential. It then produces a witty summary that frames poor prompting habits as "crimes," complete with tongue-in-cheek verdicts like "Excessive use of ‘please’ detected—your AI is being spoon-fed." The project’s creator, Devesh Sangwan, emphasizes privacy as a core principle: no uploads, no telemetry, and no external tracking.

How prompt-crimes works under the hood

The CLI leverages simple file parsing to read chat logs or exported conversations from popular AI assistants. It applies lightweight heuristics to detect common prompt anti-patterns such as:

• Overly vague instructions
• Redundant politeness tokens
• Excessive context dumps
• Improper use of system prompts

Once analysis is complete, the tool outputs a concise yet entertaining report to the terminal. Users can customize the tone by adjusting emoji density or selecting between sarcastic and supportive roast modes. All processing happens locally, eliminating cloud dependency and reducing latency.

Install in seconds with zero setup

Getting started requires just one command. Users simply run:

npx prompt-crimes

No configuration, no API keys, and no sign-up necessary. The tool auto-detects supported chat formats and scans connected directories. For those who prefer local binaries, a traditional npm install -g prompt-crimes option is also available. The lightweight design ensures it runs smoothly even on modest hardware.

Why humor matters in developer tooling

Developer-focused tools often prioritize raw functionality over user experience. By injecting humor into feedback loops, prompt-crimes aims to make learning more engaging. The satirical approach lowers the barrier to self-improvement, encouraging users to revisit their prompt engineering habits without feeling judged.

Early adopters have praised the tool for making code review and AI interaction more enjoyable. Some developers even share their roast reports in team channels as lighthearted icebreakers during stand-ups.

As AI adoption accelerates, tools like prompt-crimes highlight the growing importance of prompt literacy. The next frontier may include integrations with IDEs or version control systems, turning prompt hygiene into a visible part of the development workflow.

For developers ready to laugh—and improve—the CLI is a small but meaningful step toward better AI collaboration.