Ask HN: Good guides to securing containers used for unsupervised agentic coding

Any good guides regarding how to secure containers that will be used to run agentic AI coding assistants unsupervised?

To be specific, I'd like to use Podman containers (running maybe Debian 12) for this, and: - lock them down to not accept any incoming traffic and only allow traffic to a small set of necessary domains - prevent unauthorized access to mounted secret files, sensitive env variables, etc - monitor traffic to minimize the chance of exfiltration of any sensitive information or code - prevent any attempt to remove these protections

I've looked creating a build for this myself, but it gets more complex the more I think about it. Messing around with firewall rules, container permissions, file access, proxy servers/software, etc. starts to become hairy real quickly.

Is there a guide or pre-built images for this?