Struct CreateChildInput 

pub struct CreateChildInput {

    pub parent_session: Session,
    pub child_id: String,
    pub title: String,
    pub responsibility: String,
    pub assignment_prompt: String,
    pub subagent_type: String,
    pub workspace: String,
    pub model_override: Option<String>,
    pub model_ref_override: Option<ProviderModelRef>,
    pub runtime_metadata: HashMap<String, String>,
    pub auto_run: bool,
    pub reasoning_effort: Option<ReasoningEffort>,
    pub lifecycle: Option<String>,
    pub resident_name: Option<String>,
    pub resident_context: Option<String>,
    pub disabled_tools: Option<BTreeSet<String>>,
    pub context_fork: Option<usize>,
}

Input for creating a child session.

Fields

parent_session: Session

child_id: String

title: String

responsibility: String

assignment_prompt: String

subagent_type: String

workspace: String

Absolute path to the working directory for the child session.

model_override: Option<String>

Optional model override resolved from subagent_type routing. When None, the child inherits the parent session’s model.

model_ref_override: Option<ProviderModelRef>

Optional provider+model override resolved from subagent routing. When present, this preserves cross-provider routing for child execution.

runtime_metadata: HashMap<String, String>

Runtime metadata resolved from subagent routing (e.g. external agent config).

auto_run: bool

Whether to immediately enqueue the child for execution. Defaults to true.

reasoning_effort: Option<ReasoningEffort>

Optional reasoning effort to apply to the child’s own LLM calls. None (the default) leaves Session::reasoning_effort at None, so the provider falls back to its default. The child does NOT inherit the parent’s reasoning_effort — fan-out children that only need a quick lookup should not pay for xhigh reasoning just because the orchestrator is running at xhigh.

lifecycle: Option<String>

Lifecycle of this child: Some("resident") marks a reusable resident agent (one stable session reused for successive tasks under the same root); None/Some("oneshot") is the default throwaway child.

resident_name: Option<String>

For a resident agent, the stable reuse key (scoped to the root session).

resident_context: Option<String>

For a resident agent, how successive tasks treat prior context: "reset" (default — independent tasks) or "accumulate" (remember).

disabled_tools: Option<BTreeSet<String>>

Tool names to disable for this child (denylist; matched by EXACT ToolSchema.function.name). None (the default) = full toolset. A read-only Guardian reviewer sets e.g. {“Edit”,“Write”,“SubAgent”,…}. Carried to the child’s SpawnJob.disabled_tools via the child session metadata (see create_child_action) so the worker trims its toolset.

context_fork: Option<usize>

Model-controllable context fork (Phase 3): when Some(n) with n > 0, the last n non-system parent messages are rendered into a “Forked context from parent” block prepended to the child’s task brief. None (the default) keeps the child on a clean, freshly-seeded context.

Trait Implementations

impl Clone for CreateChildInput

fn clone(&self) -> CreateChildInput

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for CreateChildInput

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.