Hacker News Reader

This feels a lot like trying to sanitize database inputs instead of using prepared statements.

What's the equivalent of prepared statements when using AI agents?

Don't have the AI run the commands. You read them, consider them, and then run them yourself.