What would make code interpreter even move useable would be our ability to give the model a list of code libraries that are used in our code either by name or GitHub or gitlab (or any other source), and have them be referenced by the model. Current linking with the browser plugin has the plugin go on irrelevant click sprees. Providing library sources would cut unnecessary processing power and make results relevant to the tasks at hand. Thoughts?
You could make your own code interpreter type plugin to do exactly that—as long as you’re confident in your ability to sandbox the environments or are somehow comfortable with random people executing code on machines you’re responsible for.
I see your point on security. Perhaps rather than being part of code interpreter, verification of adherence to library specs would be a boon in and of itself, so maybe just a plugin to be used with a model as a type of linter.