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

# 5. Delegate and review work

> Hand one task to a specialist, follow the operation thread, and understand automatic review and escalation.

Delegation is where the AI workforce becomes operational. A source agent hands a task to an eligible specialist, the specialist works in its own team context, and the result returns to the original conversation.

## Do this

Choose **Show me** in the mission banner. The walkthrough highlights the reporting workforce, opens **Channels**, and points to the assigned Main channel or the control needed to establish one.

1. Confirm the target Team Agent has a Main channel and an active specialist.
2. In the source channel, ask SuperEmber to hand one bounded task to that specialist.
3. Open the target team's Main channel operation thread.
4. Follow the status from **Working** to **Completed**.
5. Return to the source thread and confirm the final result arrived there.

```text theme={null}
@SuperEmber delegate this to the research specialist: compare the three options in this thread and return a recommendation with risks. Do not make external changes.
```

## Ask versus delegate

| Mode     | Use it when                                                       | What the target may do                                                  |
| -------- | ----------------------------------------------------------------- | ----------------------------------------------------------------------- |
| Ask      | You need a concise read-only answer from another team's context.  | Answer the question without accepting a task handoff.                   |
| Delegate | The specialist should execute a bounded task and return a result. | Use its normal tools, subject to automatic review and human escalation. |

## What automatic review does

Before a boundary-crossing action runs, SuperEmber reviews the exact proposed call against the user's intent, authorization, effect, and risk. It can:

* approve the exact action and continue
* deny it and require a materially safer route
* pause it for a human decision

A delegated instruction is not treated as direct human authorization for destructive, permission, financial, legal, credential, or other high-impact actions.

```mermaid theme={null}
stateDiagram-v2
    [*] --> Proposed
    Proposed --> Approved: safe and authorized
    Proposed --> Denied: unsafe or outside intent
    Proposed --> HumanReview: ambiguous or high impact
    Approved --> Executed
    Denied --> SaferRoute
    HumanReview --> Executed: approved
    HumanReview --> SaferRoute: denied
```

## When a human is needed

The exact action remains paused. The target team's operation thread shows that review is needed, and the original thread receives a waiting receipt. After the decision, both receipts update to **Approved** or **Denied**.

Before approving, check the exact action, target, account, arguments, and expected effect. Approval applies once to that parked call; it is not an ongoing permission.

If the target has no Main channel, the origin thread explains that the team could not be notified. The action stays paused rather than silently executing.

## You are done when

The checklist marks **Complete a delegation** as done after a delegated task reaches `completed`.

Continue with [Try a workflow and its approvals](/onboarding/workflows-and-approvals).