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

# Modes and permissions

> Switch among default, plan, agent, and auto modes, and handle permissions and directory grants.

The mode picker under the input decides what the model may do on its own and which actions must ask you first.

## Common modes (Claude)

| UI name    | Internal value      | Behavior                                                        | Best for                            |
| ---------- | ------------------- | --------------------------------------------------------------- | ----------------------------------- |
| Default    | `default`           | Confirm reads, writes, and commands first                       | Unfamiliar repos, production config |
| Plan mode  | `plan`              | Read-only tools only; produce a plan and wait for your approval | Large changes, comparing approaches |
| Agent mode | `acceptEdits`       | Auto-accept file creates and edits                              | Day-to-day coding                   |
| Auto mode  | `bypassPermissions` | Skip permission checks                                          | Local tasks you fully trust         |

<img src="https://mintcdn.com/mossx-docs/kz4nt5KFSKzkAZwe/images/desktop/permission-dialog.png?fit=max&auto=format&n=kz4nt5KFSKzkAZwe&q=85&s=7ef448b35a40aa11e44b778fa3805a01" alt="Permission confirmation dialog" width="100%" data-path="images/desktop/permission-dialog.png" />

<Warning>
  Auto mode skips file and command confirmation. Use it only when you accept the model editing any workspace file and running commands.
</Warning>

## Permission dialog choices

| Option       | Result                                                       |
| ------------ | ------------------------------------------------------------ |
| Allow        | Approve this time                                            |
| Always allow | Do not ask again for the same kind of action in this session |
| Deny         | Skip this time                                               |

When the model asks a multiple-choice question, it uses the AskUserQuestion dialog. You can turn on system notifications in Settings.

## Plan mode and the Plan panel

Plan mode first produces a read-only plan. The app also has a **Plan panel** that lists the execution steps so you can see how far it has gone.

<img src="https://mintcdn.com/mossx-docs/kz4nt5KFSKzkAZwe/images/desktop/plan-panel.png?fit=max&auto=format&n=kz4nt5KFSKzkAZwe&q=85&s=1590d7c14e0f810128783dc163a30850" alt="Plan panel" width="100%" data-path="images/desktop/plan-panel.png" />

## Directory Grant

If Claude Read / Bash fails because the path is outside the allowed working directory, an inline Directory Grant card may appear:

* Approve writes a session-level grant. The next launch injects `--add-dir`
* Deny leaves the failure closed

This is a different layer of safety from the mode picker.

## Codex sandbox and collaboration mode

Codex has separate sandbox / collaboration mode settings. You can combine them with chat permission modes: one side controls “ask you or not,” the other controls which directories and network the command can actually touch.
