Moving theorem proving upstream into compilers, sandboxes, and the browser seems like the future when we're dealing with increasingly sophisticated AIs. I'm working on similar formal methods but applied to agent sandboxing; do you see Z3 as a better fit than lean? https://github.com/coproduct-opensource/nucleus
I'm building Nucleus for exactly this problem - using information flow control and formal methods, we can prevent confused deputies by proofs instead of heuristics.
Everyone agrees that agent security is an area that needs significant improvement, and quickly. Using methods from information flow control, this is a lightweight demo of how web search can taint a Claude session so it doesn't allow writing after a accessing untrusted data.
This can be configured via profiles to more more or less restrictive.
Treat this as an example for now, more to come in the future.
Thanks! Currently network identity is host-based, but in the middle of introducing SPIFFE based on ZTunnel. Should be done in the next couple of days.
Runtime enforcement means that any side effects are routed through a proxy (nucleus-tool-proxy) that does realtime checks on permissions and gates the behavior.
SPIFFE for MicroVM agents is a compelling idea and I'll update when this is ready.