feat(agents): persist evaluation gate failures as system experiences for cross-conversation learning

将评估门控的失败教训自动沉淀为全局系统经验，实现跨对话学习闭环：

1. KnowledgeClientService 新增 saveExperience() 方法
   - 封装 knowledge-service 的 POST /memory/experience API
   - 支持保存 experienceType、scenario、content、confidence 等字段
   - 错误处理：失败时返回 null，不影响主流程

2. CoordinatorAgentService 评估门控回调扩展
   - Gate 失败(RETRY/SUPPLEMENT)时，异步将失败教训写入 knowledge-service
   - fire-and-forget 模式：不阻塞重试流程，失败仅打印警告
   - 经验格式：scenario = "阶段 + 规则名"，content = "具体要求描述"
   - experienceType = KNOWLEDGE_GAP，confidence = 45（低于默认50，需积累）

3. 学习闭环机制
   - 经验写入后状态为 PENDING，需 admin 审核激活（isActive=true）
   - 激活后 Context Injector 的 searchExperiences() 语义匹配检索
   - 未来对话中注入 prompt → AI 第一次就遵循规则，避免重复犯错
   - knowledge-service 自动去重：0.9 相似度阈值合并，confidence +5/次

4. 零侵入保证
   - 无评估规则时 Gate 直接 PASS → 不保存任何经验 → 原有流程零影响
   - agent-loop.ts 未修改，所有逻辑在 coordinator 回调闭包内完成

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

packages/services/conversation-service/src/infrastructure/agents/coordinator/coordinator-agent.service.ts

@@ -219,8 +219,9 @@ export class CoordinatorAgentService implements OnModuleInit {
         turnCount: number,
         agentsUsedInLoop: string[],
       ) => {
-        return this.evaluationGate.evaluate(null, {
-          stage: context.consultingState?.currentStageId || null,
+        const stage = context.consultingState?.currentStageId || null;
+        const gateResult = await this.evaluationGate.evaluate(null, {
+          stage,
           collectedInfo: context.consultingState?.collectedInfo || null,
           assessmentResult: context.consultingState?.assessmentResult || null,
           responseText,
@@ -229,6 +230,22 @@ export class CoordinatorAgentService implements OnModuleInit {
           hasConverted: false,
           agentsUsed: agentsUsedInLoop,
         });
+
+        // Gate 失败时，将失败教训异步保存为系统经验（fire-and-forget）
+        if (
+          !gateResult.passed &&
+          (gateResult.action === 'RETRY' || gateResult.action === 'SUPPLEMENT')
+        ) {
+          this.persistGateFailureAsExperience(
+            gateResult,
+            stage,
+            context.conversationId,
+          ).catch(err =>
+            this.logger.warn(`Failed to persist gate failure experience: ${err}`),
+          );
+        }
+
+        return gateResult;
       };
 
       // 6. Build agent loop params
@@ -501,6 +518,44 @@ export class CoordinatorAgentService implements OnModuleInit {
     }
   }
 
+  // ============================================================
+  // Gate Failure → System Experience
+  // ============================================================
+
+  /**
+   * 将评估门控失败沉淀为系统经验
+   * 经验写入 knowledge-service，状态为 PENDING，需 admin 审核后激活
+   * 激活后，未来对话中 Context Injector 的 searchExperiences() 会检索到并注入 prompt
+   * 从而让 AI 在第一次生成回复时就遵循人工经验，避免重复犯错
+   */
+  private async persistGateFailureAsExperience(
+    gateResult: import('./evaluation-gate.service').GateResult,
+    stage: string | null,
+    conversationId: string,
+  ): Promise<void> {
+    const failedRules = gateResult.results.filter(r => !r.passed);
+    if (failedRules.length === 0) return;
+
+    const stageLabel = stage || '通用';
+
+    for (const rule of failedRules) {
+      const scenario = `${stageLabel}阶段 — ${rule.ruleName}`;
+      const content = `在${stageLabel}阶段，${rule.message || rule.ruleName + '检查未通过'}。AI 回复前应确保满足此要求，避免生成不合格内容后重试。`;
+
+      await this.knowledgeClient.saveExperience({
+        experienceType: 'KNOWLEDGE_GAP',
+        scenario,
+        content,
+        sourceConversationId: conversationId,
+        confidence: 45,
+      });
+
+      this.logger.debug(
+        `Persisted gate failure as experience: [${rule.ruleType}] ${scenario}`,
+      );
+    }
+  }
+
   // ============================================================
   // Event Mapping
   // ============================================================

packages/services/conversation-service/src/infrastructure/knowledge/knowledge-client.service.ts

@@ -287,6 +287,38 @@ export class KnowledgeClientService implements OnModuleInit {
     }
   }
 
+  /**
+   * 保存系统经验
+   * 用于将评估门控的失败教训沉淀为全局经验，影响未来所有对话
+   */
+  async saveExperience(params: {
+    experienceType: string;
+    content: string;
+    scenario: string;
+    sourceConversationId: string;
+    confidence?: number;
+    relatedCategory?: string;
+  }): Promise<SystemExperience | null> {
+    try {
+      const response = await fetch(`${this.baseUrl}/api/v1/memory/experience`, {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify(params),
+      });
+
+      if (!response.ok) {
+        console.error(`[KnowledgeClient] saveExperience failed: ${response.status}`);
+        return null;
+      }
+
+      const data = (await response.json()) as ApiResponse<SystemExperience>;
+      return data.success ? data.data : null;
+    } catch (error) {
+      console.error('[KnowledgeClient] saveExperience error:', error);
+      return null;
+    }
+  }
+
   // ============================================================
   // Convenience Methods (for Specialist Agents)
   // ============================================================