import type z from 'zod';
import type { AgentLLMIntegrationParam } from '../Integrations/types.js';
import type { IMCPClient } from '../interfaces.mcp.js';
import type { IPermissionManager, PermissionManagerContext } from '../interfaces.permission.manager.js';
import type { OtelInfoType } from '../types.js';
import type { AgentEventStreamer } from './stream/types.js';
import type { AgentMessage, AgentOutputBuilder, AgentToolDefinition, CreateArvoAgentParam } from './types.js';
/**
 * The Core Cognitive Loop of the Arvo Agent.
 *
 * This function implements the **ReAct (Reason + Act)** pattern, orchestrating the interactive
 * session between the Large Language Model (LLM) and the available Tool ecosystem.
 *
 * @remarks
 * **Hybrid Execution Strategy:**
 * The loop handles two types of tool executions differently:
 * 1. **Synchronous Tools (Internal & MCP):** These are executed **immediately** within the loop.
 *    The results are added to the history, and the LLM is called again in the same tick.
 * 2. **Asynchronous Tools (Arvo Services):** These interrupt the loop. The function returns
 *    the tool call definition, signaling the parent `ArvoResumable` to **emit an event and suspend**.
 *
 * **Self-Correction:**
 * If the LLM's final output fails the Contract's Output Schema validation (via `outputBuilder`),
 * the loop catches the error and feeds it back to the LLM for auto-correction.
 *
 * @param param.currentSubject - The Arvo orchestration subject of the currently executing agent instance.
 *   Passed through to every internal tool invocation so tools can identify which agent is calling them.
 * @param param.parentSubject - The Arvo orchestration subject of the parent orchestrator that invoked
 *   this agent. `null` when this agent is the root of the orchestration tree. Also forwarded to
 *   every internal tool invocation.
 */
export declare const agentLoop: (param: {
    currentSubject: string;
    parentSubject: string | null;
    initLifecycle: AgentLLMIntegrationParam["lifecycle"];
    system: string | null;
    messages: AgentMessage[];
    tools: AgentToolDefinition[];
    outputFormat: z.ZodTypeAny;
    outputBuilder: AgentOutputBuilder;
    llmResponseType: NonNullable<NonNullable<CreateArvoAgentParam["inferenceConfig"]>["responseType"]>;
    llm: NonNullable<NonNullable<CreateArvoAgentParam["inferenceConfig"]>["llm"]>;
    preInferenceHook: NonNullable<NonNullable<NonNullable<CreateArvoAgentParam["inferenceConfig"]>["hooks"]>["preInference"]> | null;
    postInferenceHook: NonNullable<NonNullable<NonNullable<CreateArvoAgentParam["inferenceConfig"]>["hooks"]>["postInference"]> | null;
    mcp: IMCPClient | null;
    agentCycles: {
        current: number;
        max: number;
    };
    currentTotalExecutionUnits: number;
    currentTotalUsageTokens: {
        prompt: number;
        completion: number;
    };
    onStream: AgentEventStreamer;
    permissionPolicy: string[];
    permissionManager: IPermissionManager | null;
    permissionManagerContext: PermissionManagerContext;
}, config: {
    otelInfo: OtelInfoType;
}) => Promise<{
    messages: {
        content: {
            type: "tool_result";
            toolUseId: string;
            content: string;
        } | {
            type: "text";
            content: string;
        } | {
            type: "media";
            content: string;
            contentType: {
                type: "image";
                name: string;
                mediatype: string;
                format: "base64";
            } | {
                type: "file";
                name: string;
                mediatype: string;
                format: "base64";
            };
        } | {
            type: "tool_use";
            toolUseId: string;
            name: string;
            input: Record<string, any>;
        };
        role: "user" | "assistant";
        seenCount: number;
    }[];
    toolCalls: {
        type: "tool_use";
        toolUseId: string;
        name: string;
        input: Record<string, any>;
    }[];
    agentCycles: {
        current: number;
        max: number;
    };
    executionUnits: number;
    tokenUsage: {
        prompt: number;
        completion: number;
    };
    output?: undefined;
} | {
    messages: {
        content: {
            type: "tool_result";
            toolUseId: string;
            content: string;
        } | {
            type: "text";
            content: string;
        } | {
            type: "media";
            content: string;
            contentType: {
                type: "image";
                name: string;
                mediatype: string;
                format: "base64";
            } | {
                type: "file";
                name: string;
                mediatype: string;
                format: "base64";
            };
        } | {
            type: "tool_use";
            toolUseId: string;
            name: string;
            input: Record<string, any>;
        };
        role: "user" | "assistant";
        seenCount: number;
    }[];
    output: {
        [x: string]: any;
    } & {
        __id?: import("arvo-core").CreateArvoEvent<Record<string, unknown>, string>["id"];
        __executionunits?: import("arvo-core").CreateArvoEvent<Record<string, unknown>, string>["executionunits"];
    };
    agentCycles: {
        current: number;
        max: number;
    };
    executionUnits: number;
    tokenUsage: {
        prompt: number;
        completion: number;
    };
    toolCalls?: undefined;
}>;
//# sourceMappingURL=agentLoop.d.ts.map