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

# Inbox and approvals

> Read Issue notifications and resolve agent permission requests.

**Inbox** brings together Issue activity that needs your attention, including assignment changes, review updates, failed Runs, and approval requests. Open a notification to inspect its context, then use **Open** to go to the [Issue](https://docs.longoz.ai/issues.md).

## Manage notifications

Selecting a notification marks it read. You can use **Mark as unread** to keep it visible for later, or **Mark all as read** to clear the unread count.

Search Inbox and use its display settings to show read items, place unread items first, or change ordering.

The notification menu can remove all current items, read items, or items for completed Issues. These actions dismiss notifications from your Inbox. They do not delete the Issues or their history. New notifications can still arrive afterward.

On an Issue, use **Subscribe** to receive attention for later activity. **Unsubscribe** stops subscription-based notifications, although responsibilities such as assignment can still cause activity to reach your Inbox.

## Resolve a permission request

Reading or dismissing a notification does not approve the requested operation. An approval notification can show a permission card directly in its preview. Permission cards also appear in the Issue's context. **Approval Inbox** separates **Pending** requests from **Recent** decisions.

Before deciding, review the operation, target, working context, and risk indicators. Sensitive details are masked, and previews can be shortened. If you need more context, open the Issue and inspect the current Run.

When you have permission to resolve the request, choose:

* **Allow once** to permit this request.
* **Always allow** to save an exact-match trust rule as well as allow the request.
* **Decline** to deny the request.

Always allow is scoped to the same Project, agent, Computer, provider, tool, risk tier, and matching arguments. It is not blanket access to a tool or Computer. Reuse also requires the rule creator to retain owner or admin authority, and does not apply to Strict Runs.

If the request is an **Agent question**, select the offered answers or enter a response where allowed, then choose **Send answer**. Questions do not offer Always allow.

## Check that work continued

A saved decision can still be waiting for delivery to the Computer or for the provider to apply it. Read the card's lifecycle message before assuming the agent has continued. An allowed tool can still fail after the decision is applied.

Expired, withdrawn, or superseded requests cannot be resolved. If the request changes before your decision, Longoz reloads it. Review the updated request before trying again. If you can read a card but cannot resolve it, ask an organization administrator for the required access.

For failed or blocked work, open the Issue and use the actions available for its current Run. A recovery notification is a prompt to inspect that work; marking it read does not resume the agent.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.