Feature hasn't been suggested before.
Describe the enhancement you want to request
Here’s my idea: place restrictions on the AI agent regarding terminal (shell) commands.
For instance, block certain commands or require explicit user confirmation before execution—especially for commands that delete folders or files and could impact the project.
I use OpenCode to check my code structure, and I once told it something like "Delete everything"—let's just say I gave a vague instruction like that—and the agent took it so literally that it deleted almost all my code.
Sure, it was my fault for not clarifying what I meant by "delete," but it would be great to add command restrictions so the agent doesn't just execute whatever it wants—if you catch my drift.
We could block specific command types or require user approval for critical actions (like deletion).
Especially since it uses commands that are irreversible:
Remove-Item -LiteralPath "ExampleRoute" -Recurse -Force
You know what I mean?
Another idea is to provide context for each project type—for example, telling it: "This project is a landing page using this specific API, so you shouldn't do X or Y here."
And one more idea—why not?
Why not support other package managers instead of always relying on NPM? We could use tools like Bun or Pnpm, especially since NPM isn't great at detecting packages containing viruses or malware—and the number of malicious packages being uploaded is on the rise.
Thanks for listening, guys! :P
If the text looks a bit weird or "cursed," it's because I used Google Translate—I'm from Argentina ;-;
Feature hasn't been suggested before.
Describe the enhancement you want to request
Here’s my idea: place restrictions on the AI agent regarding terminal (shell) commands.
For instance, block certain commands or require explicit user confirmation before execution—especially for commands that delete folders or files and could impact the project.
I use OpenCode to check my code structure, and I once told it something like "Delete everything"—let's just say I gave a vague instruction like that—and the agent took it so literally that it deleted almost all my code.
Sure, it was my fault for not clarifying what I meant by "delete," but it would be great to add command restrictions so the agent doesn't just execute whatever it wants—if you catch my drift.
We could block specific command types or require user approval for critical actions (like deletion).
Especially since it uses commands that are irreversible:
Remove-Item -LiteralPath "ExampleRoute" -Recurse -ForceYou know what I mean?
Another idea is to provide context for each project type—for example, telling it: "This project is a landing page using this specific API, so you shouldn't do X or Y here."
And one more idea—why not?
Why not support other package managers instead of always relying on NPM? We could use tools like Bun or Pnpm, especially since NPM isn't great at detecting packages containing viruses or malware—and the number of malicious packages being uploaded is on the rise.
Thanks for listening, guys! :P
If the text looks a bit weird or "cursed," it's because I used Google Translate—I'm from Argentina ;-;