Struct SpawnAgentTool 

pub struct SpawnAgentTool { /* private fields */ }

The spawn_agent tool. It is registered on StaticToolRegistry and shared across sessions of the owning AgentCore via process_tools (it is not a process-global singleton — a single process may host multiple AgentCore instances, each with its own copy). At construction time it captures everything needed to run a nested turn, because ToolContext only carries cwd/fs/shell/http/cancel/current_model, not the provider registry, policy, or tool set.

Implementations

impl SpawnAgentTool

pub fn new( profiles: Arc<BTreeMap<String, SubagentProfile>>, registry: Arc<ProviderRegistry>, policy: Arc<dyn SandboxPolicy>, process_tools: Arc<dyn ToolRegistry>, base_prompt: Option<String>, ) -> Self

Constructs a spawn_agent tool. When profiles is empty, the caller should not register this tool (the profile enum in the schema will be an empty set, so calls will always fail) — see Self::has_profiles.

pub fn has_profiles(profiles: &BTreeMap<String, SubagentProfile>) -> bool

Whether any profiles were discovered. The assembler uses this to decide whether to register this tool.

Trait Implementations

impl Tool for SpawnAgentTool

fn schema(&self) -> &ToolSchema

Tool metadata. Returns a reference to avoid allocating on every call.

fn safety_hint(&self, _args: &Value) -> SafetyClass

Provides a safety-level hint to the sandbox policy without actually executing the tool.

fn describe<'a>( &'a self, args: &'a Value, _ctx: ToolContext<'a>, ) -> BoxFuture<'a, ToolCallDescription>

Generates a “self-description” before execution, for display to the ACP client.

fn execute(&self, args: Value, ctx: ToolContext<'_>) -> ToolStream

Initiates a tool call and returns an event stream.