fix(agents): filter out internal strategy notes from followUp output

The model was putting internal notes like "引导回移民话题" in the followUp
field instead of actual user-facing questions. Two fixes:

1. Schema: describe followUp as "必须以？结尾，禁止填写内部策略备注"
2. agent-loop: only yield followUp if it contains ？or ? (question mark)

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

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

@@ -346,7 +346,8 @@ export async function* agentLoop(
 
           yield { type: 'text', content: answer, timestamp: Date.now() };
 
-          if (parsed.followUp) {
+          // followUp 必须是面向用户的问题（含？），过滤掉内部策略备注
+          if (parsed.followUp && /？|\?/.test(parsed.followUp)) {
             const followUp = smartTruncate(parsed.followUp, MAX_FOLLOWUP_LENGTH);
             yield { type: 'text', content: '\n\n' + followUp, timestamp: Date.now() };
           }

packages/services/conversation-service/src/infrastructure/agents/schemas/coordinator-response.schema.ts

@@ -18,7 +18,7 @@ export const CoordinatorResponseSchema = z.object({
     'casual_chat',            // 闲聊/打招呼："你好"
   ]),
   answer: z.string().describe('直接回答用户的文本，简洁精准，默认100字以内'),
-  followUp: z.string().optional().describe('跟进引导问题（可选）'),
+  followUp: z.string().optional().describe('直接对用户提出的跟进问题（必须以？结尾，必须是用户能看到的自然语言问题，禁止填写内部策略备注）'),
 });
 
 export type CoordinatorResponse = z.infer<typeof CoordinatorResponseSchema>;