State why a command is run outside the sandbox

[joshspicer]

Testing #303798

This is a bit confusing for me. Is this curl command run outside the sandbox because auto approval is denied for curl. If so, why can't I just approve it to run in the sandbox. Should we be ignoring this auto approval rules when in the sandbox?

My first thought was that this popped up because I hadn't configured the network allow list, but reading it again I don't think that is relevant here.

I wasn't given an option to augment the sandbox config with this untrusted URL. It would be nice to configure that here instead of needing to dig through settings

[joshspicer]

another example that confuses me. Looks like I needed to update the sandbox.network setting.

    "chat.tools.terminal.sandbox.network": {
        "allowedDomains": ["spicer.dev"],
        "deniedDomains": [],
        "allowTrustedDomains": false
    }

[alexdima]

Is this curl command run outside the sandbox because auto approval is denied for curl. If so, why can't I just approve it to run in the sandbox. Should we be ignoring this auto approval rules when in the sandbox?

The LLM is aware ahead of time (via prompting) what domains you have allowed / blocked. It then thinks and instead of running a command which will fail inside the sandbox due to domains being blocked, it will offer to run it outside the sandbox.

---

I acknowledge the feedback. You would expect the UX would give you a straight-forward way to add to the list of allowed domains and then proceed to run the command inside the sandbox.

[alexdima]

Let's use #304053 to track that we should have UX to give users straight-forward way to add to the list of allowed domains and then proceed to run the command inside the sandbox

Let's use this issue to improve the "why did the LLM decide to run this command outside the sandbox"

[alexdima]

We will now show the reason filled in by the LLM: