ISPO SDK
    Preparing search index...

    Interface AgentSpawnInput

    Markdown

    Spawn an agent terminal at a target (spec §10.22). Gated PER TARGET:

    • SELF-RUN — projectId is THIS project → needs only agent.dispatch. Use it to launch and auto-run your own worker agent (e.g. an eval runner). No privileged grant, no consent prompt.
    • ORCHESTRATION — projectId is another project → needs the privileged agent.orchestrate grant, and the first spawn into each target is confirmed by the user once. The spawned agent gets the TARGET's authority, never this project's (no authority laundering). autoSubmit fires the seed prompt unattended (honored for self-run and orchestration alike — §10.22). Pair the returned terminalId with host.navigate({ kind: 'agents', terminalId }) to bring the user straight to the spawned terminal.
    interface AgentSpawnInput {
        agent: "claude" | "codex" | "opencode" | "pi" | "cursor" | "ispo";
        autoSubmit?: boolean;
        projectId: string;
        seedPrompt?: string;
        timeoutMs?: number;
    }
    Index
    agent: "claude" | "codex" | "opencode" | "pi" | "cursor" | "ispo"
    autoSubmit?: boolean

    Fire the staged prompt automatically. Honored for both self-run and cross-target orchestration (§10.22); only the grant-less MCP agent binding stays prepare-only.

    projectId: string
    seedPrompt?: string

    Seed prompt staged into the new terminal. 1..64KB.

    timeoutMs?: number

    Client RPC budget for this INTERACTIVE call (ms). A cross-target spawn waits on the first-use consent prompt the user must answer, plus the agent boot — which exceeds the default 10s and would otherwise time out while the host completes the spawn (orphaning the agent). Pass a larger value (capped by the transport). Host-ignored; consumed by the SDK transport.