Struct SubAgentRunner 

pub struct SubAgentRunner {
    pub work_dir: PathBuf,
    pub source_root: PathBuf,
    pub llm_client: Arc<dyn LlmClient>,
    pub event_tx: Option<UnboundedSender<AgentEvent>>,
    pub override_llm_client: Option<Arc<dyn LlmClient>>,
}

Runs a subagent to completion.

The subagent gets its own AgentLoop with its own context window. It runs the given task prompt and returns the result. Subagents cannot spawn other subagents — the spawn_subagent tool is intentionally excluded.

Fields

work_dir: PathBuf

source_root: PathBuf

llm_client: Arc<dyn LlmClient>

event_tx: Option<UnboundedSender<AgentEvent>>

override_llm_client: Option<Arc<dyn LlmClient>>

Optional override LLM client (for model override).

Implementations

impl SubAgentRunner

pub fn new( work_dir: PathBuf, source_root: PathBuf, llm_client: Arc<dyn LlmClient>, ) -> Self

pub fn with_event_sink(self, tx: UnboundedSender<AgentEvent>) -> Self

pub fn with_override_llm_client(self, client: Arc<dyn LlmClient>) -> Self

pub async fn run( &self, def: &SubAgentDef, task_prompt: &str, ) -> SdkResult<SubAgentResult>

Run a subagent with the given definition and task prompt.

Returns the subagent’s result including final content and token usage.

pub fn run_background( &self, def: SubAgentDef, task_prompt: String, ) -> JoinHandle<SdkResult<SubAgentResult>>

Run a subagent in the background, returning a handle to await the result.