[{"data":1,"prerenderedAt":16},["ShallowReactive",2],{"casestudies":3},[4],{"path":5,"title":6,"description":7,"number":8,"date":9,"readingTime":10,"topics":11},"\u002Fcasestudies\u002Fsandboxed-evaluation","Sandboxed evaluation","Run model-written Python inside a throwaway microVM across several turns, read each result back, then delete the VM.","001","2026-08-19","8 min",[12,13,14,15],"evals","agents","ssh","python sdk",1787592506928]