How to whitelist commands and handle unsafe input in sys.exec?

[gunnsth]

Question

I’m pretty new to gptscript and have a question about safely executing commands with sys.exec (or similar features). Is there a built-in way to:

1. Whitelist Commands:

   • Allow only specific commands to be executed, either by binary name (e.g., ls) or absolute path (e.g., /bin/ls)?
   • Raise an error or ignore commands that are not explicitly allowed?

2. Filter Unsafe Input:

   • Prevent execution of inputs containing special shell symbols like \``, |, $, ;, or &&`, which could spawn additional processes?
   • For example, how would the library handle inputs like ls -al $(pwd) or echo $(rm -rf /) or ls -al `pwd` etc.?
   • Can users provide a custom validation or sanitization function for input arguments to enforce stricter security?

3. Safety Guarantees:

   • Are there existing mechanisms in the library to ensure these concerns are handled safely, or would these need to be implemented by the user?

---

Context

I want to ensure that commands executed through the library are secure and controlled, particularly in scenarios where user-provided input is involved.

Running commands through firejail or something like that could also make sense to have a completely controlled environment (at least on linux).

Any guidance on how to approach this would be greatly appreciated. Thanks for creating such an awesome tool!