Struct LaunchRequest 

pub struct LaunchRequest<'a> {
    pub path: &'a str,
    pub task: Option<&'a str>,
    pub stdin_is_terminal: &'a dyn Fn() -> bool,
    pub model: Option<String>,
    pub workdir: &'a str,
    pub yolo: bool,
    pub allow: Vec<String>,
    pub max_depth: Option<usize>,
    pub regions: HashMap<String, String>,
    pub no_seed_commands: bool,
    pub output_request: Option<OutputSpec>,
}

What lev run was asked for, before any of it is resolved.

One struct because these are one thing: the command line. Each field is a flag the user typed, and grouping them keeps the difference between “what was asked for” and “what that resolves to” visible - resolve_spawn_args turns this into a SpawnArgs, and the two are deliberately different types.

Fields

path: &'a str

The blueprint path or name, as given.

task: Option<&'a str>

The task text, if it was given rather than read from stdin or an editor.

stdin_is_terminal: &'a dyn Fn() -> bool

Whether stdin is a terminal, injected so the editor path is testable.

model: Option<String>

--model, overriding the blueprint’s choice.

workdir: &'a str

The working directory tools run in.

yolo: bool

--yolo: run unattended.

allow: Vec<String>

--allow: tools permitted outright.

max_depth: Option<usize>

--max-depth: sub-agent tree cap.

regions: HashMap<String, String>

--<region> seeds, keyed by caller-input region name.

no_seed_commands: bool

--no-seed-commands: refuse the blueprint’s command seeds.

output_request: Option<OutputSpec>

The output shape the caller asked for, overriding the blueprint’s.