pub fn handle_subagent_tool(
params: Value,
task_config: TaskConfig,
sub_recipes: HashMap<String, SubRecipe>,
working_dir: PathBuf,
cancellation_token: Option<CancellationToken>,
) -> ToolCallResultExpand description
Note: SubRecipe.sequential_when_repeated is surfaced as a hint in the tool description (e.g., “[run sequentially, not in parallel]”) but not enforced. The LLM controls sequencing by making sequential vs parallel tool calls.