Terminal Command Limits

[mredig]

Can we get some control over which terminal commands the agent can use?

My specific issue is that occasionally the agent will try to use sed instead of the built in zed editor tool. This is far more destructive than the wonderful system you designed to allow line by line approvals and edits after the ai makes changes.

I'd like to add sed to a disallow list (or at least something that will prompt permission to proceed) when it runs terminal commands as it would have very few legit uses of the command for me.

In regards to existing discussions, #32101 seems to be more related to claude code usage than the zed agent, plus it deals more with default values while I'm asking for customization. Ideally, I'd like my request to apply to both, but I'm really only asking for zed agent compliance.