@runloop/agent-axon-client
    Preparing search index...

    Type Alias AgentDefinition

    Definition for a custom subagent that can be invoked via the Agent tool.

    type AgentDefinition = {
        background?: boolean;
        criticalSystemReminder_EXPERIMENTAL?: string;
        description: string;
        disallowedTools?: string[];
        effort?: "low" | "medium" | "high" | "max" | number;
        initialPrompt?: string;
        maxTurns?: number;
        mcpServers?: AgentMcpServerSpec[];
        memory?: "user" | "project" | "local";
        model?: string;
        permissionMode?: PermissionMode;
        prompt: string;
        skills?: string[];
        tools?: string[];
    }
    Index

    Claude SDK

    background? criticalSystemReminder_EXPERIMENTAL? description disallowedTools? effort? initialPrompt? maxTurns? mcpServers? memory? model? permissionMode? prompt skills? tools?

    Claude SDK

    background?: boolean

    Run this agent as a background task (non-blocking, fire-and-forget) when invoked

    criticalSystemReminder_EXPERIMENTAL?: string

    Experimental: Critical reminder added to system prompt

    description: string

    Natural language description of when to use this agent

    disallowedTools?: string[]

    Array of tool names to explicitly disallow for this agent

    effort?: "low" | "medium" | "high" | "max" | number

    Reasoning effort level for this agent. Either a named level or an integer

    initialPrompt?: string

    Auto-submitted as the first user turn when this agent is the main thread agent. Slash commands are processed. Prepended to any user-provided prompt.

    maxTurns?: number

    Maximum number of agentic turns (API round-trips) before stopping

    mcpServers?: AgentMcpServerSpec[]
    memory?: "user" | "project" | "local"

    Scope for auto-loading agent memory files. 'user' - ~/.claude/agent-memory//, 'project' - .claude/agent-memory//, 'local' - .claude/agent-memory-local//

    model?: string

    Model alias (e.g. 'sonnet', 'opus', 'haiku') or full model ID (e.g. 'claude-opus-4-5'). If omitted or 'inherit', uses the main model

    permissionMode?: PermissionMode

    Permission mode controlling how tool executions are handled

    prompt: string

    The agent's system prompt

    skills?: string[]

    Array of skill names to preload into the agent context

    tools?: string[]

    Array of allowed tool names. If omitted, inherits all tools from parent

    Settings

    Member Visibility

    On This Page

    Claude SDK