Command sandbox and approvals

While DataMind Oda works on a task, it sometimes runs commands on your computer. This page covers how the chat's access mode confines those commands, what a blocked command looks like, and how you allow one wider retry. For how commands run and how to read their output, see Shell commands.

What a command may change

The chat's access mode decides where a command may change files on your computer:

• In Read Only, a command that would change files is stopped.
• In Workspace Write, a command may change files inside the workspace folder; a change outside it is stopped. This is the default.
• In Full access, a command runs without these limits.

To change the mode, click the mode button in the message box or type /permission; Workspaces, files and permissions covers the modes and what each one allows.

When a command is blocked

When a command tries to change a file the mode does not allow, the file sandbox stops that change:

• The command's result carries a short sandbox note that names the mode the command ran under, in this form: [sandbox: file access denied under read-only mode].
• The denied change is not made.
• The note also carries an escalation hint: the exact same command can be retried once with wider access, and you are asked before it runs.
• Oda reports the denial and either asks you for that retry, or continues with the task another way.

The same sandbox stops Oda's file changes; File safety and error messages covers the file side.

Allowing a wider retry

After a blocked command, Oda may ask to run the exact same command once more, with wider access:

1. Oda asks you first, with a one-sentence reason. Nothing runs before your answer.
2. The retry repeats the exact same command, once, with wider access only: Oda asks for the narrowest wider access that can work. When Oda needs more access shows how far each mode can widen.
3. If you allow it, that one step runs with the wider access, and the result reports the mode it actually ran under.
4. If you reject it, or the request is cancelled, the command does not run, and Oda continues with the task another way.

• A retry can only widen. A request that does not go past the mode the command ran under is refused, and a command that stays inside the current mode needs no request.
• The widening covers that one step only: the chat keeps its access mode, and the next command that goes wider asks you again.
• A rejection ends that retry: Oda does not run the widened command on its own.

The request works like Oda's other approval cards; Your first task shows the choices and what each answer does.

When the limits cannot be set up

On some computers the file limits cannot be applied to the folder, for example when Windows denies the permission change they need. Oda does not drop the limits quietly. It reports what went wrong:

DataMind Oda can't limit writes to <folder> because Windows denied changing its permissions.

Before running the command without restrictions, it asks:

DataMind Oda can't limit writes to <folder>. Run this command without restrictions?

• If you allow it, exactly that one command runs without restrictions; nothing else changes.
• If you decline, the command does not run: Running this command without restrictions was not approved, so it did not run. The ways forward are to allow it when the command is retried, to switch the chat to Full access where it is offered, or to use a folder you own.
• With approval prompts turned off, the request is refused without asking you, and the command does not run.
• When no approval prompt can be shown at all, the command does not run either: No approval prompt is available here, so this command did not run. The ways forward then shorten to a folder you own.

Limits worth knowing

• In the sandbox modes, programs cannot open named pipes, so a command that needs one fails.
• If the sandbox itself fails to start or supervise the command, the result says the command did not run and that this is a sandbox problem, not a command failure. Oda treats it that way and does not report it as the command's own failure.
• A background job belongs to the chat that started it; another chat cannot read it. Shell commands covers reading and stopping jobs.

Questions, answered

What does the sandbox note in a command's result mean?

The command tried to change a file the chat's mode does not allow, and the file sandbox stopped that change. The note names the mode the command ran under, and the denied change is not made.

Why did my command report that file access was denied?

It tried to change files outside the limits of the chat's mode. You can allow the wider retry Oda asks for, change the chat's access mode, or use a folder you own.

Can a blocked command be retried more than once?

One retry of the exact same command, and only wider. A request that does not go past the mode the command ran under is refused.

Does allowing a wider retry change the mode for the rest of the chat?

No. The widening covers that one step; the chat keeps its access mode, and the next command that goes wider asks you again.

What happens if I reject a wider retry?

The command does not run, and Oda continues with the task another way. The ways forward are to allow it when the command is retried, to switch the chat to Full access where it is offered, or to use a folder you own.

What happens when no approval prompt can be shown?

Nothing is approved without you: the command does not run. With approval prompts turned off, the request is refused without asking you.

Why could Oda not apply the limits to my folder?

On some computers the limits need a permission change that Windows denies, so they cannot be applied to that folder. Oda does not drop the limits quietly: it says why, and asks before running one command without restrictions.

Why do some commands fail while the limits are on?

In the sandbox modes, programs cannot open named pipes, so a command that needs one fails. It is the sandbox working as intended, not a fault in the command.

What does it mean when the result says the sandbox runner failed?

The sandbox itself could not start or supervise the command, so the command did not run. The result says this is a sandbox problem, not a command failure, and Oda treats it that way.