/**
 * Example: End-to-End Memory + Darwin Integration
 *
 * Closed-loop in three lines:
 *   1. Run an agent.
 *   2. After multi-critic, persist findings via McpMemoryBridge.
 *   3. Before the next run, fetch relevant lessons and inject them.
 *
 * Backend-agnostic: defaults to `@studiomeyer/local-memory-mcp` so it works
 * out of the box with no API keys. Override `localMemory()` with
 * `remoteMemory(url, …)` to point at any MCP-compliant memory server.
 *
 * Run demo (requires `@studiomeyer/local-memory-mcp` reachable via npx):
 *   npx tsx examples/memory-darwin-integration.ts
 *
 * The demo deliberately uses a stub agent runner so it doesn't depend on
 * a live LLM connection. Replace `simulateAgentRun()` with your real
 * `runAgent()` call once you wire this into your codebase.
 */
import { type Lesson, type RetrievableFeedbackStore } from './mcp-memory-bridge.js';
import type { MultiCriticResult } from '../src/evolution/multi-critic.js';
/**
 * Render a list of past lessons as a context block that you can prepend to
 * the next agent's system prompt. Empty lessons → empty string (cheap to
 * call unconditionally).
 */
export declare function renderLessonContext(lessons: Lesson[], maxChars?: number): string;
export interface AgentRunInput {
    agentName: string;
    topic: string;
    /** Optional context block (e.g. rendered past lessons). */
    context?: string;
}
export interface AgentRunOutput {
    text: string;
    critic: MultiCriticResult;
}
/**
 * The orchestration shape. Bring your own implementations for the two
 * function fields when wiring this into a real codebase:
 *   - runner: anything that takes (topic, context) → {text, critic}
 *   - store:  the McpMemoryBridge (or any RetrievableFeedbackStore)
 */
export interface ClosedLoopOptions {
    runner: (input: AgentRunInput) => Promise<AgentRunOutput>;
    store: RetrievableFeedbackStore;
    /** Cap how many lessons to inject. Default: 5. */
    fetchLimit?: number;
    /** Override persist thresholds (same shape as persistFeedback options). */
    persistThresholds?: {
        lowThreshold?: number;
        highThreshold?: number;
        minOutputChars?: number;
    };
}
/** Run one closed-loop turn: fetch lessons → run → persist findings. */
export declare function runClosedLoopTurn(input: AgentRunInput, options: ClosedLoopOptions): Promise<{
    run: AgentRunOutput;
    lessonsUsed: number;
    persisted: boolean;
    reason: string;
}>;
//# sourceMappingURL=memory-darwin-integration.d.ts.map