Enum AgentInput 

pub enum AgentInput {
    Text(String),
    Message(Vec<ContentBlock>),
    Resume {
        continuation: Box<ContinuationEnvelope>,
        tool_call_id: String,
        confirmed: bool,
        rejection_reason: Option<String>,
    },
    SubmitToolResults {
        continuation: Box<ContinuationEnvelope>,
        results: Vec<ExternalToolResult>,
    },
    Continue,
}

Input to start or resume an agent run.

Variants

Text(String)

Start a new conversation with user text.

Message(Vec<ContentBlock>)

Start a new conversation with rich content (text, images, documents).

Resume

Resume after a confirmation decision.

Fields

continuation: Box<ContinuationEnvelope>

The versioned continuation envelope from AwaitingConfirmation.

tool_call_id: String

ID of the tool call being confirmed/rejected.

confirmed: bool

Whether the user confirmed the action.

rejection_reason: Option<String>

Optional reason if rejected.

SubmitToolResults

Resume after external tool execution.

Use this after TurnOutcome::PendingToolCalls when ToolRuntime::External is set. The caller must provide a result for every pending tool call listed in the continuation.

The SDK validates the continuation envelope version, appends the tool results to the message store, and continues to the next LLM turn.

Fields

continuation: Box<ContinuationEnvelope>

The versioned continuation from TurnOutcome::PendingToolCalls.

results: Vec<ExternalToolResult>

One result per pending tool call. The order does not matter, but every tool_call_id from the continuation must be covered.

Continue

Continue to the next turn (for single-turn mode).

Use this after TurnOutcome::NeedsMoreTurns to execute the next turn. The message history already contains tool results from the previous turn.

Trait Implementations

impl Debug for AgentInput

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

Formats the value using the given formatter.