Comment by Tyriar
1 month ago
It's not just the .vscode folder though, the Python extension for example executes code in order to provide language services. How could this threat detection possibly be complete? In this new LLM-assisted world a malicious repository could be as innocuous as a plain text prompt injection attack hidden in a markdown file, or some random command/script that seems like it could be legitimate. There are other mitigations in place and in progress to help with the LLM issue, but it's a hard problem.
This demonstrates the actual real-world problem, though. You're saying "this is a complex problem so I'm going to punt and depend on the user to resolve it". But in real life, the user doesn't even know as much as you do about how Code and its plugins interact with their environment. Knowledgewise, most users are not in a good position to evaluate the dangers. And even those who could understand the implications are concentrating on their goal of the moment and won't be thinking deeply about it.
You're relying the wrong people, and at the wrong time, for this to be very effective.
> It's not just the .vscode folder though, the Python extension for example executes code in order to provide language services.
Which code? Its own Code (which the user already trusts anyway), or code from the workspace (automatically)? My expectation with a language-server is that it never code from the workspace in a way which could result in a side effect outside the server gaining understanding about the code. So this makes little sense?
Your expectation is wrong in this case for almost all languages. The design of Pylance (as is sorta forced by Python itself) chooses to execute Python to discover things like the Python version, and the Python startup process can run arbitrary code through mechanisms like sitecustomize.py or having a Python interpreter checked into the repo itself. To my knowledge, Go is one of the few ecosystems that treats it as a security failure to execute user-supplied code during analysis tasks, many languages have macros or dynamic features that basically require executing some amount of the code being analyzed.