Am I overthinking AI security, or is this actually useful? Need your feedback.

Hey everyone 👋

I've been building a zero-trust local firewall for AI coding agents (like Cursor, Claude Code, etc.).

The problem I’m trying to solve is that LLMs constantly hallucinate package names. Bad actors are starting to register those fake packages with malicious code (slopsquatting). So if your AI agent autonomously runs `npm install` or `pip install` on a hallucinated package, your machine gets compromised instantly.

I just spent weeks updating my tool (Agentinel) to natively intercept terminal commands across JS, Python, and Rust in under 1 millisecond. It uses a massive offline vulnerability database to block the AI from pulling malware before the shell even executes it.

But as a solo dev, I sometimes get tunnel vision. So I wanted to ask the community directly:

Do you guys think a local firewall for AI agents is actually useful right now?

If yes, would you install it locally? And if not, why? Do you just rely on Docker, or do you think the threat is overblown?

I’d love some brutally honest feedback before I push this further!