Your question is Secure LLM Sandbox for Untrusted Code. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How would you implement a secure, isolated sandbox execution environment to run and test untrusted code generated by an LLM?