Struct SubAgentTool 

pub struct SubAgentTool<State: Send + Sync, Ctx: Send + Sync = ()> { /* private fields */ }

A Tool adapter that exposes a SubAgent to a parent agent — the surface that turns “agents calling agents” into an ordinary tool call.

When the parent model calls this tool, SubAgentTool runs the wrapped sub-agent as a child run at the configured parent_depth and returns the child’s final assistant text as the crate::harness::tool::ToolResult content. This makes an entire agent composable as a single tool call, so a model orchestrating tools is, transparently, a model orchestrating models.

Because the Tool trait gives call no access to the live parent crate::harness::context::RunContext, the depth the child runs at is fixed at construction (parent_depth, default 0). Nesting deeper sub-agents is expressed by constructing the inner tool with a larger parent_depth. For fully context-threaded invocation (reading the live parent depth) call SubAgent::invoke_in_parent directly instead of going through the tool.

Implementations

impl<State: Send + Sync, Ctx: Send + Sync> SubAgentTool<State, Ctx>

pub fn new(subagent: Arc<SubAgent<State, Ctx>>) -> Self

Wraps subagent as a tool invoked at parent_depth = 0 (child runs at depth 1). The tool name defaults to the sub-agent name.

pub fn with_tool_name(self, name: impl Into<String>) -> Self

Overrides the model-visible tool name.

pub fn with_parent_depth(self, parent_depth: usize) -> Self

Sets the caller depth this tool invokes the child at; the child runs at parent_depth + 1. Use this to express deeper nesting through the tool path (where the live parent depth is not available).

pub fn with_parameters(self, parameters: Value) -> Self

Overrides the model-visible JSON Schema for the tool arguments.

Trait Implementations

impl<State, Ctx> Tool<State> for SubAgentTool<State, Ctx>

where State: Send + Sync, Ctx: Send + Sync + Default,

fn name(&self) -> &str

Canonical tool name.

fn description(&self) -> &str

Human/model readable description.

fn schema(&self) -> ToolSchema

Returns the model-visible schema for this tool.

fn call<'life0, 'life1, 'async_trait>( &'life0 self, state: &'life1 State, call: ToolCall, ) -> Pin<Box<dyn Future<Output = Result<ToolResult>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Executes the tool against application state and a validated call.

fn call_with_context<'life0, 'life1, 'async_trait>( &'life0 self, state: &'life1 State, call: ToolCall, context: ToolExecutionContext, ) -> Pin<Box<dyn Future<Output = Result<ToolResult>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Executes the tool with access to caller run context.

fn policy(&self) -> ToolPolicy

Returns the SDK-owned safety/runtime policy for this tool.

fn display_label(&self, _call: &ToolCall) -> Option<String>

Returns the human-facing label for this specific call.

fn display_detail(&self, call: &ToolCall) -> Option<String>

Returns the human-facing detail for this specific call.

fn timeout_policy(&self, _call: &ToolCall) -> ToolTimeout

Returns the invocation timeout behavior for this specific call.