Containing the Blast Radius: Hardening WSL2 for AI Coding Agents
The LiteLLM supply chain attack exposed a risk most developers running AI agents haven't thought through: a compromised dependency doesn't need to escalate privileges — it already has your source code, SSH keys, and an unobstructed path to the internet. Four layers of hardening for local AI agent workloads on WSL2: interop lockdown, scoped filesystem access via bind mounts, VS Code remote isolation over SSH, and uid-based outbound traffic restriction via nftables and Squid.