diff --git a/docs/BUG_HISTORY.md b/docs/BUG_HISTORY.md
index 22bbd43c..70ea2bca 100644
--- a/docs/BUG_HISTORY.md
+++ b/docs/BUG_HISTORY.md
@@ -1489,3 +1489,17 @@
 - 防复发：普通聊天的可见措辞不得由技术 packet 或事件台账驱动；内部收敛状态只能记录和计算，不能成为 Agent 每轮必须复述的脚本。
 - 相关记录：BUG-074、BUG-075、BUG-078、BUG-079
 - 修复版本：本次修复提交
+
+## BUG-081 | 开放叙事只共情不继续引导
+
+- 状态：resolved
+- 首次发现：2026-07-26
+- 最近更新：2026-07-26
+- 影响面：生时校正开放叙事的中间轮回复
+- 用户现象：用户讲完一段包含明确行动与转折的经历后，Agent 只分析感受和性格，没有自然询问后续经历，对话停在原地。
+- 根因：为消除问卷式强制追问，中间轮 prompt 将提问设为完全可选，且校验器只检查已有问题是否可见，没有要求事件轮承担继续引导叙事的职责。
+- 修复：普通中间轮 prompt 要求在本轮新事件后以一个贴着上下文的开放问题收尾；优先询问事件之后的下一步，继续禁止机械补年月、结果或转折。若模型仍只共情不引导，服务端直接补一句开放问题，不增加第二次模型调用；无模型可用时的安全兜底也保留自然后续问题。
+- 验证：回归覆盖“研究院实习后因价值观冲突辞职”场景，确认只共情不引导的回复会保留原文并补上后续问题，且不触发模型重试。
+- 防复发：开放叙事不等于被动倾听；事件轮默认负责自然推进，但不得恢复固定问卷模板。
+- 相关记录：BUG-075、BUG-080
+- 修复版本：本次修复提交
diff --git a/frontend/src/lib/conversational-rectification/narrative-agent.ts b/frontend/src/lib/conversational-rectification/narrative-agent.ts
index e21e0c4e..d085c00e 100644
--- a/frontend/src/lib/conversational-rectification/narrative-agent.ts
+++ b/frontend/src/lib/conversational-rectification/narrative-agent.ts
@@ -582,7 +582,7 @@ function promptFor(
       rules: [
         "优先回应用户话里的真实内容、感受、选择或转折，不要复述成档案摘要。",
         "除非用户主动询问校时进度或技术结果，不要说已记录、先记为、对校时有价值、参与候选时间核对、不会因单一事件确认分钟，也不要主动谈候选范围、分盘、评分、收敛或内部处理。",
-        "不必每轮提问。需要提问时只问自然推进对话真正需要的问题，不要机械补年月、结果或转折。",
+        "用户刚提供实际经历时，回应之后必须用一个贴着上下文的开放问题收尾，自然引导他继续往后讲；优先问这件事之后发生了什么、接下来去了哪里或做了什么，不要机械补年月、结果或转折。",
         "只返回 narrative 和 evidenceRequest。没有在 narrative 中逐字提出一个用户可见的明确问题时，evidenceRequest 必须为 null；若提出问题，evidenceRequest.prompt 必须与 narrative 中的问题文字完全一致。followUp 只表达 kind、answerMode 和 proposedDate，不要生成 evidenceId，目标事件由服务器绑定。",
       ],
       retryIssues: boundedReceiptIssues(retryIssues),
@@ -639,7 +639,7 @@ function promptFor(
 function fallbackNarrative(packet: RectificationTechnicalPacket, phase: RectificationNarrativePhase): string {
   const candidate = packet.candidate;
   if (phase === "intermediate") {
-    return "我听到了。你可以顺着这段经历继续说，也可以自然讲下一件想到的事。";
+    return "我听到了。你愿意接着说说这件事之后发生了什么吗？";
   }
   const phaseLine = phase === "final" && candidate.status === "ready_for_confirmation"
     ? "当前证据已形成候选总结，但仍有残余不确定性；只有明确确认后才会替换当前排盘时间。"
@@ -812,11 +812,21 @@ export async function generateRectificationNarrative(input: {
       ), { signal, attempt });
       const modelId = modelIdSchema.parse(generated.modelId ?? defaultModelId);
       const parsedOutput = parseModelOutput(generated.text, input.packet, input.context ?? {});
-      const output = input.phase === "intermediate"
+      const withoutHiddenFollowUp = input.phase === "intermediate"
         && parsedOutput.evidenceRequest
         && !/[?？]/u.test(parsedOutput.narrative)
         ? { ...parsedOutput, evidenceRequest: null }
         : parsedOutput;
+      const output = input.phase === "intermediate"
+        && input.context?.latestEvidence?.length
+        && !technicalDiscussionRequestPattern.test(input.context.latestUserText ?? "")
+        && !/[?？]/u.test(withoutHiddenFollowUp.narrative)
+        ? {
+            ...withoutHiddenFollowUp,
+            narrative: `${withoutHiddenFollowUp.narrative.trim()}\n\n这件事之后，紧接着发生了什么？`,
+            evidenceRequest: null,
+          }
+        : withoutHiddenFollowUp;
       const validation = validateNarrativeAgainstPacket(
         output,
         input.packet,
diff --git a/frontend/tests/conversational-narrative-agent.test.ts b/frontend/tests/conversational-narrative-agent.test.ts
index 2db931e7..126234db 100644
--- a/frontend/tests/conversational-narrative-agent.test.ts
+++ b/frontend/tests/conversational-narrative-agent.test.ts
@@ -455,17 +455,43 @@ test("passes the user's latest words to an ordinary intermediate reply", async (
       }],
     },
     generator: generator([{
-      narrative: "第一次长期离开家去工作，适应过程应该不轻松。你可以接着讲。",
+      narrative: "第一次长期离开家去工作，适应过程应该不轻松。后来最先发生了什么？",
       evidenceRequest: null,
     }], prompts),
   });
 
   assert.match(prompts[0] ?? "", /离开家乡去上海开始第一份长期工作/);
   assert.match(prompts[0] ?? "", /不要把自己写成记录员/);
+  assert.match(prompts[0] ?? "", /必须用一个贴着上下文的开放问题收尾/);
   assert.doesNotMatch(prompts[0] ?? "", /latestEvidence/);
   assert.doesNotMatch(prompts[0] ?? "", /候选范围.*05:16/);
 });
 
+test("adds a lightweight guide when an event reply only empathizes", async () => {
+  const prompts: string[] = [];
+  const result = await generateRectificationNarrative({
+    phase: "intermediate",
+    packet: syntheticTechnicalPacket(),
+    context: {
+      latestUserText: "实习到10月份，发现有人抢项目骗经费，我就辞职了",
+      latestEvidence: [{
+        dateLabel: "2020-10",
+        summary: "因价值观冲突辞职",
+        domain: "career",
+      }],
+    },
+    generator: generator([{
+      narrative: "这段经历确实很冲击，也能理解你为什么不愿意继续留下。",
+      evidenceRequest: null,
+    }], prompts),
+  });
+
+  assert.equal(result.attempts, 1);
+  assert.equal(result.fallbackUsed, false);
+  assert.match(result.narrative, /这件事之后，紧接着发生了什么？$/);
+  assert.equal(prompts.length, 1);
+});
+
 test("keeps internal event domains and suggested-domain routing out of the narrator context", async () => {
   const prompts: string[] = [];
   await generateRectificationNarrative({
diff --git a/frontend/tests/conversational-rectification-orchestrator.test.ts b/frontend/tests/conversational-rectification-orchestrator.test.ts
index 10c98a1c..1cc3814d 100644
--- a/frontend/tests/conversational-rectification-orchestrator.test.ts
+++ b/frontend/tests/conversational-rectification-orchestrator.test.ts
@@ -1693,7 +1693,7 @@ test("a rejected intermediate narrative falls back without discarding the event"
 
   const stored = value.cases.get(startActionId)?.row;
   assert.equal(turn.turnVersion, 1);
-  assert.match(turn.narrative, /顺着这段经历继续说|自然讲下一件/);
+  assert.match(turn.narrative, /这件事之后.*发生了什么/);
   assert.equal(stored?.privateCandidate.resultId, null);
   assert.equal(stored?.eventEvidence.length, 1);
   assert.equal(stored?.validationReceipts.length, 2);
@@ -1989,7 +1989,7 @@ test("vague, future, and unmatched answers stay conversational and never score",
   }
 });
 
-test("a free Agent reply clears the previous structured question instead of inheriting it", async () => {
+test("a free Agent reply clears the previous structured question and adds an open guide", async () => {
   const value = harness({
     readyAfterEvidenceCount: 99,
     continueLatestEvent: true,
@@ -2010,7 +2010,7 @@ test("a free Agent reply clears the previous structured question instead of inhe
 
   assert.equal(second.evidenceRequest, null);
   assert.match(second.narrative, /顺着这段经历继续说|自然讲下一件/);
-  assert.doesNotMatch(second.narrative, /[？?]/);
+  assert.match(second.narrative, /这件事之后，紧接着发生了什么？/);
 });
 
 test("a non-scoring packet failure responds to the current turn instead of replaying the prior agent message", async () => {
@@ -2140,7 +2140,7 @@ test("a failed regenerate saves a fallback turn without changing evidence, candi
 
   const storedAfter = value.cases.get(startActionId)?.row;
   assert.equal(regenerated.turnVersion, answered.turnVersion + 1);
-  assert.match(regenerated.narrative, /顺着这段经历继续说|自然讲下一件/);
+  assert.match(regenerated.narrative, /这件事之后.*发生了什么/);
   assert.deepEqual(storedAfter?.eventEvidence, evidenceBefore);
   assert.deepEqual(storedAfter?.privateCandidate, candidateBefore);
   assert.equal(value.counts().reserveCount, countsBefore.reserveCount);