> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zerotwo.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Rules and approvals

ZeroTwo uses permission modes, sandboxing, and approval prompts to control what agents can do on your machine. Use these controls instead of a separate CLI rules language.

## Permission modes

In the ZeroTwo desktop app, choose how much access ZeroTwo has by default:

* **Workspace access** — read and edit files in the active project or Cowork folder
* **Auto-review** — ZeroTwo can automatically review requests for additional access
* **Full access** — edit files more broadly and run networked commands with fewer prompts (use carefully)

See [Permission modes](/permission-modes) and [Permissions](/permissions) for details.

## Sandboxing

Command execution and file tools run inside ZeroTwo's sandbox model. Escalations that leave the sandbox prompt for approval unless your settings allow them.

* [Sandboxing](/sandboxing)
* [Auto-review](/sandboxing/auto-review)
* [Agent approvals](/agent-approvals-security)

## Approvals during a run

When ZeroTwo needs something outside the current policy — network, a path outside the workspace, or a sensitive command — it asks before continuing. You can approve once, allow for the session, or deny.

Smart / auto-review settings can reduce repeated prompts for similar safe requests. Review suggested escalations carefully.

## Project instructions vs permissions

| Mechanism                                                    | Use it for                                          |
| ------------------------------------------------------------ | --------------------------------------------------- |
| [`AGENTS.md` / `ZEROTWO.md`](/agent-configuration/agents-md) | Repo norms, preferred commands, what *not* to touch |
| Permission + sandbox settings                                | What ZeroTwo is *allowed* to do on the machine      |
| Approvals                                                    | One-off escalations during a run                    |

## Windows

On Windows, also see [Windows sandbox](/windows/windows-sandbox) and [WSL](/windows/wsl).

## Related

* [Permissions](/permissions)
* [Sandboxing](/sandboxing)
* [AGENTS.md](/agent-configuration/agents-md)
* [Subagents](/agent-configuration/subagents)