Why no "Auto-Approve All"?

[twardoch]

I’m running Roo in VSCode on macOS, on a user account that has very strictly limited rights. The entire user account is its own sandbox, it doesn’t have system admin privileges.

Roo is very useful with Claude Code as provider. I was able to make it run agentically for hours — except when the agent is trying to run some shell command. I mean, it needs approval for stuff like cd .. && ls -la, which is ridiculous.

I don’t need to approve it — I can always roll back to an earlier git state, to a backup, or easily reinstate the entire user account.

So — how can I make Roo NOT ASK FOR ANY APPROVAL?

The Cursor app agent, Claude Code CLI, Gemini CLI, Codex CLI — they all have a "yolo"/"auto-approve all" mode. But Roo’s UI is complicated, extremely clunky huge UI for approval.

Please don’t put patronizing hurdles in front of users.

[mrubens]

Go into the auto approve settings and add *

[mrubens]

Agree in general about the clunky UI. We should clean that up. The intention isn’t to be patronizing through - there are real security risks for people who aren’t as sandboxed as you are.

[twardoch]

Agree in general about the clunky UI. We should clean that up. The intention isn’t to be patronizing through - there are real security risks for people who aren’t as sandboxed as you are.

The fact that there are options for approval is not patronizing. But the fact that among many GUI options the "Auto-Approve All" is not listed but instead is hidden — that's patronizing. It’s the classic case of software developers thinking they’re smarter than the users of their software. Which for a product aimed for developers (albeit vastly different ones) is, well, patronizing :)

[mrubens]

Not completely hidden right? Isn’t there copy in the settings saying to add * to approve all commands? Or did that get lost recently?

[mrubens]

In any case, you’re probably the first person to tell me that some advanced settings are too subtle. Achievement unlocked 😂

[mrubens]

(I do agree that it’s probably worth making “allow everything” more prominent - thank you for the feedback there)

[dosubot[bot]]

Hi @twardoch! I'm Dosu and I’m helping the Roo-Code team.

Roo Code does support an "auto-approve all" mode for shell commands, but it’s hidden behind some strong warnings. You can enable this via the Auto-Approve Toolbar: click the toolbar above the chat input, use the master toggle to enable all permissions, and in the settings panel, look for the "Always approve allowed execute operations" option. To truly allow everything, add * as a command prefix—this will auto-approve all shell commands, but it’s considered very risky and is not recommended unless you’re absolutely sure about the safety of your environment and workflow [docs].

This setup is similar to the "yolo" modes in Cursor, Claude Code CLI, etc., but Roo’s UI is more granular and explicit about the risks. The approval UI is intentionally prominent because running shell commands agentically is a major security vector, but if you have strong rollback/backup strategies and are comfortable with the risk, you can bypass all prompts as described above.

If you want more details or screenshots, check the official docs. If you run into any quirks with the auto-approve menu, note that there have been recent bug fixes in this area [release notes].

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}   

[twardoch]

OK, thanks, so it’s just a case of patronizing UI. :)