Many developers run @kirodotdev's agent directly on their local machine so it can see the same filesystem, toolchain, and localhost services they do. That's what makes it effective for building and testing real projects. Today, developers manage that risk by reviewing every command before approving it. As agents take on longer tasks and developers run more sessions in parallel, that attention becomes the bottleneck. Local sandboxing changes the trust model: even after a command is approved, the operating system still limits what it can read and write on your machine, and which network destinations it can reach. The agent still runs natively on your machine, but the commands it executes are restricted by the kernel. You can turn sandboxing on or off mid-session without restarting. Available now on macOS, Linux, and Windows: kiro.dev/blog/local-sandboxi…