When using Cursor’s agent mode, I’ve observed that it doesn’t seem to recognize .env files. While I strongly agree with the principle of not exposing the contents of .env files to the agent for security reasons, this behavior presents a challenge when attempting to have the agent write code that interacts with these files.
Specifically, if I try to prompt the agent to write code that involves reading an .env file, Cursor often misinterprets this. It might assume the file path is incorrect or that the .env file hasn’t been created yet. This has led to undesirable actions, such as the agent autonomously modifying the README file, citing a lack of permissions to create the .env file.
My request is simple: I would greatly appreciate it if Cursor could be updated to recognize the existence and path of .env files without necessarily reading their contents. This would prevent the agent from making incorrect assumptions and taking unprompted actions, while still maintaining the crucial privacy of my environment variables.
I encountered the same behavior. Though the model doesnt seem to be trying harder to read it or write to it.