VSCode's SSH Agent Is Bananas (2025) (fly.io)

A recent announcement highlights the integration of VSCode with remote editing capabilities over SSH, emphasizing its growing popularity in the software development community, especially among those using LLMs for code generation. This integration aims to enhance LLM-generated code by creating a closed-loop system where the LLM generates code, the resulting errors are fed back to it for improvements, and this process iterates continuously. This approach mitigates “hallucinations,” or inaccuracies, often present in LLM-generated output by allowing developers to test in a controlled environment. However, the method used by VSCode raises concerns regarding security and safety. Unlike Emacs' TRAMP, which operates within the constraints of the remote environment, VSCode implements a more extensive system by downloading an agent that can perform various operations over SSH. This includes file system navigation, file editing, and executing shell commands, all while establishing a WebSockets connection back to the VSCode frontend. Such capabilities could pose risks, especially when used on development or production servers, prompting caution among users who might deploy this setup in sensitive environments.