Sandman grew out of the problem of running code you did not write and would like to continue not fully trusting. That usually means containers, limits, tokens, logs, and a long list of things that are obvious only after they have already gone sideways once.

The related write-up on designing a system to run untrusted code gets into the shape of the problem. Sandman is the project version: take the threat model seriously, keep the developer experience tolerable, and do not confuse “it ran on my machine” with a security story.