sandbox
A sandbox restricts the resources a process can access.
Use this when Untrusted code or an agent must run with limited access.
without the term
“I want the agent to run code but I'm scared it'll delete stuff or hit the internet. Can you run it somewhere where it can only touch this folder and nothing else?”
with the term · How do you ask a model for sandbox?
Run this workload in a sandbox with only its required files and network access. Verify normal execution and denial outside those boundaries.
What is sandbox?
The process can reach only the files, network hosts, and system calls its configuration allows. What it actually blocks depends on that configuration, so test both allowed and denied actions.
Why does sandbox matter when vibe coding?
Instructions cannot substitute for an enforced execution boundary.
How do you do sandbox?
Choose the allowed resources and verify both intended operations and blocked ones.
What goes wrong with sandbox?
Disabling the entire boundary to work around one missing permission.