---
title: ".delegate()"
description: "Expose a sealed child agent as a model-visible delegation tool."
---

> Documentation Index
> Fetch the complete documentation index at: https://xsaf.ilha.build/llms.txt
> Use this file to discover all available pages before exploring further.

# .delegate()

## Signature

```ts
agent.delegate(agent: XsafAgent, options?: DelegateOptions): XsafAgent
```

## Seal and register a child

```ts
const researcher = agent({
  ...researchConfig,
  name: "researcher",
  description: "Research one focused question",
});

const parent = agent(parentConfig).sandbox(sandbox).delegate(researcher, {
  passContext: false,
  approval: "human",
});
```

The child appears to the parent model as a tool accepting `{ prompt: string }`. Its name must be lowercase snake_case and cannot collide with a local or MCP tool. The derived child session ID is `<parent-session>:delegate:<delegate-name>`.

## Context boundary

`passContext` defaults to `false`: the child receives the delegated prompt but not the parent's history, tools, memory driver, sandbox permissions, or channel metadata. Set it to `true` only when the child needs the parent's messages. The child always keeps its own model, tools, resources, and runtime configuration.

## Execution policy

Delegates support `"auto"`, `"human"`, or callback approval. Execution uses the delegate-specific `sandbox` option when supplied, otherwise the parent's default sandbox. Executable delegates require an applicable explicit sandbox.

Starting the parent starts its children. Shutdown closes them through the normal reverse-order lifecycle. The parent emits `delegate.started` and `delegate.completed`; completion is emitted even when the child request fails so telemetry can close the span without exposing the delegated prompt.

Registering a child seals its configuration. See [Tools & Security](/recipes/tools) for shared execution policy.

Source: https://xsaf.ilha.build/xsaf/delegate/index.mdx