Struct SubAgentCallInput 

pub struct SubAgentCallInput {
    pub agent: Option<String>,
    pub input: Option<String>,
    pub code_review: bool,
    pub continue_session: bool,
    pub run_in_background: bool,
}

Input for calling a sub-agent.

The same visible tool dispatches to TWO implementations: an external ACP agent harness when agent is provided, or the internal sub-agent (a nested harness) when agent is omitted or empty.

Fields

agent: Option<String>

The agent harness to call (e.g. “gemini”). Must be one of the supported agents listed in the tool description.

Optional: if omitted (or empty), an internal agent runs the task instead — a fresh nested harness with an empty context, in auto-approve mode, that persists nothing and returns only its final report.

input: Option<String>

The message to send to the sub-agent.

Optional: if omitted (or empty), the last message sent to a sub-agent in this session is reused automatically. If no sub-agent has been called yet, an error is returned telling the caller to provide one.

code_review: bool

Whether this sub-agent task is a CODE REVIEW.

When true, the harness appends a contract to the prompt: the sub-agent’s FINAL REPORT must start with an HTML comment header declaring the review outcome — <!-- severity: green -->, <!-- severity: yellow --> (minor issues / bad practice at most) or <!-- severity: red --> (something critical was found). The header is consumed by the client: it is never rendered, it only tints the sub-agent box (green/yellow/red). Without the flag (or without a header in the report) the box keeps its neutral per-agent color.

Optional; defaults to false.

continue_session: bool

Resume the agent’s most recent session, keeping its context (default).

true (or omitted): the call resumes the sub-agent’s most recent ACP session for this agent (tracked per agent name in the calling session), keeping its conversation context — the token-efficient default for review/iteration loops. false: start a brand-new session with clean context, for tasks unrelated to the previous one (the --continue/-c CLI convention, inverted to a positive flag).

Graceful degradation: a harness without session-resume support, or one whose stored session no longer exists (e.g. it restarted and lost state), silently falls back to a fresh session/new.

Optional; defaults to true. Ignored when agent is omitted (the internal agent never persists state across calls).

run_in_background: bool

Run the sub-agent in the BACKGROUND (default false).

true: the harness spawns the sub-agent without blocking, returns a task_id immediately, and the final report is delivered later — as an automated completion notification in a subsequent turn. Query progress at any time with the subagent_status tool (by task_id, or omit its argument to list all background tasks).

Omitted or false: the call blocks until the sub-agent finishes and returns its report directly (the default synchronous behavior).

Trait Implementations

impl Clone for SubAgentCallInput

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SubAgentCallInput

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for SubAgentCallInput

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for SubAgentCallInput

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.