Every AI coding assistant has the same fundamental problem: it can touch any file it wants on your system.
Whether you're using Cursor, Claude Code, Cline, Aider, or any other agentic tool, the moment you point it at your project folder, it has free rein. And sometimes, it will make a mistake.
The tools don't solve this for you
- Cursor: Has a .cursorrules file, but that's just instructions. The agent can still edit anything.
- Claude Code: Has permission modes, but "plan mode" still lets the agent see all your files.
- Aider: Has a --read-only flag, but then it can't actually do work.
None of these give you a hard boundary. They're all "please be careful" rather than "you cannot do this."
The fix that works for every tool
It doesn't matter which AI tool you use. The fix is the same: give the agent its own drive letter.
Create a folder for the task, map it to a drive, point the tool at that drive:
subst W: C:i-workspaces\my-project
Now the agent's entire working universe is W:\. It can't see your .env, your node_modules cache, your SSH keys, or your other projects. It literally doesn't exist from the agent's perspective.
This works with every AI tool because they all just read and write files. They don't know they're on a mapped drive.
Why I use ScsSubst
The built-in Windows subst command works, but mappings don't survive reboots. I use ScsSubst (https://subst.scsoi.com/) because it:
- Makes persistent mappings that survive reboots
- Has a simple GUI to manage all your drives
- Shows disk usage at a glance
- One-time $9.99 purchase, no subscription
The point
Stop trying to make the AI "be careful." Make it impossible for it to cause damage.
An isolated drive isn't a feature — it's the baseline. Every AI agent on your system should be working inside its own drive.
Top comments (0)