fix(consult): a stored smalltalk reply is not a prior answer; summary counts as one (acceptance fix for BUG-1071)

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0199rbQDTsUbCVw84wc8BTFe

docs/BUG_HISTORY.md

@@ -14395,8 +14395,8 @@
 - 现象：会话里第二个问题起（「那我爸呢」「我妈是不是不太在意我」），回答仍是口语开场加四个二级标题的完整解读，重述上一轮已经写过的宫位；用户感受是「太宽泛」。
 - 触发条件：同一会话已有一轮本命解读后的任何追问。
 - 根因：`jyotishInstructions`「Call run-jyotish-consultation before answering every turn, including short follow-ups」、`natalSpokenReportContract` 末句「One natal … question still uses the opener-plus-skeleton above」、每轮用户轮的 `natalAnswerShapeInstruction()`「一次写完这四个二级标题」三处都不区分首轮与追问轮。
-- 修复：`consultation-thinking-plan.ts` 新增 `followUpAnswerShapeInstruction()`（≤ 200 字、第一句答问的那句、是非题先答是 / 不是、最多三处盘面事实、不写二级标题、不写行动清单、不重述上一轮、不解释用户为什么问；明确换领域或要求完整看才回首轮形状）、`hasPriorAssistantAnswer(history)`（存量历史里已有非空 assistant 正文即追问轮，不看问句字面）与 `natalUserTurnShape()`（daily 不变 / 追问 / 首轮）；`route.ts` 用户轮按它切指令，grounded 兜底同样切换（同一份 `legacyNatalInstruction`）；思考栏与工具上下文带 `followUpTurn`。合同末句改为「首轮用开场 + 骨架，追问轮按用户轮里的指令」。工具仍每轮调用，扣点与结算不变（D7）。
-- 验证：新文件 `frontend/tests/consult-answer-the-question-20260927.test.ts`：真实 Agent + 记录提示词的假模型 + 公开盘 golden，追问轮 → 写作步的提示词含「这是追问轮 / 不超过 200 字」、不含三标题指令，工具跑 1 次、模型 2 次，无标题短正文（< 160 字）整段放出并结算扣点 1 次；首轮对照用例保持开场 + 三标题。`consultation-thinking-plan.test.ts` 覆盖 `hasPriorAssistantAnswer` / `natalUserTurnShape` 四种情况（空历史、只有用户轮、空白 assistant、有回答；daily 不受历史影响）。
+- 修复：`consultation-thinking-plan.ts` 新增 `followUpAnswerShapeInstruction()`（≤ 200 字、第一句答问的那句、是非题先答是 / 不是、最多三处盘面事实、不写二级标题、不写行动清单、不重述上一轮、不解释用户为什么问；明确换领域或要求完整看才回首轮形状）、`hasPriorAssistantAnswer(history, { summaryText })`（存量历史里已有非空、且不是寒暄回复（`responseKind: "smalltalk"`）的 assistant 正文即追问轮，会话摘要存在也算；不看问句字面。Claude 验收补修：子代理首版只看 role，「你好」的寒暄回复也存成 assistant 消息，会把其后第一个正式问题当追问；历史窗口现在把 `responseKind` 透传到 tail）与 `natalUserTurnShape()`（daily 不变 / 追问 / 首轮）；`route.ts` 用户轮按它切指令，grounded 兜底同样切换（同一份 `legacyNatalInstruction`）；思考栏与工具上下文带 `followUpTurn`。合同末句改为「首轮用开场 + 骨架，追问轮按用户轮里的指令」。工具仍每轮调用，扣点与结算不变（D7）。
+- 验证：新文件 `frontend/tests/consult-answer-the-question-20260927.test.ts`：真实 Agent + 记录提示词的假模型 + 公开盘 golden，追问轮 → 写作步的提示词含「这是追问轮 / 不超过 200 字」、不含三标题指令，工具跑 1 次、模型 2 次，无标题短正文（< 160 字）整段放出并结算扣点 1 次；首轮对照用例保持开场 + 三标题。`consultation-thinking-plan.test.ts` 覆盖 `hasPriorAssistantAnswer` / `natalUserTurnShape` 四种情况（空历史、只有用户轮、空白 assistant、有回答；daily 不受历史影响）。验收补修加：寒暄回复后仍是首轮形状、摘要存在即追问、空白摘要不算；`consultation-session-history.test.ts` 加 `responseKind` 透传用例。
 - 防复发：D6 硬红线——追问判定只用历史事实，禁正则识别意图（沿用 BUG-976/977）；route 合同测试断言 `natalUserTurnShape({ entrypoint: consultEntrypoint, history })` 与 `followUpTurn` 的接线。
 - 相关记录：BUG-1070、BUG-1073、BUG-1053（答案由看过工具结果的步写）、BUG-976/977（禁正则）。
 - 复发自：无

docs/tasks/PROGRESS-consult-answer-the-question-20260927.md

@@ -74,3 +74,22 @@
 - `next build` 默认 Turbopack 对软链的 `node_modules` 报 `Symlink … points out of the filesystem root`（与 `PROGRESS-consult-gender-optional-20260927` 同一现象），改用 `next build --webpack` 量 `/` 状态与首屏 gzip；数字口径与该记录一致（rootMainFiles，gzip level 9）。
 - lint 有 127 条既有 warning（0 error），不在本单范围，未动。
 - 真实模型输出未验（无凭据）：见 `docs/testing/consult-answer-the-question-20260927.md`。
+
+## Claude 验收补修（2026-09-27 晚）
+
+- 发现：`hasPriorAssistantAnswer` 只看 `role === "assistant"`，而寒暄快路（BUG-976/977）的回复也以 `responseKind: "smalltalk"` 存成 assistant 消息（`route.ts` `p_response_message`），历史窗口又把 `responseKind` 丢掉。后果：用户先说「你好」再问第一个正式问题，会被当成追问轮（≤ 200 字、无骨架）。会话摘要把早先轮次折掉后 tail 里可能没有 assistant，则反过来误判为首轮。
+- 修：`consultation-session-history.ts` 的 `StoredTurn` / `ConsultationHistoryMessage` 带可选 `responseKind`，窗口 tail 透传；`hasPriorAssistantAnswer(history, { summaryText })` 忽略寒暄回复、摘要存在即追问；`natalUserTurnShape` 与 `route.ts` 两处调用带 `historyWindow.summaryText`。`product-voice.ts` Never 段一句改措辞（原句第二个分句读成也被禁止）。
+- 断言三栏：`consult-single-pass-answer-20260927.test.ts`（route 源码匹配：原值不带 summaryText / 新值带 / 原因如上）、`consultation-voice-contract.test.ts`（Never 句：原值分号并列 / 新值拆两句 / 原因措辞）。新增用例见 `consultation-thinking-plan.test.ts`、`consultation-session-history.test.ts`。
+- 复跑门禁数字见下方「Claude 验收」。
+
+## Claude 验收（2026-09-27 晚，Node 22.14，本工作树，含补修）
+
+| 项 | 结果 |
+| --- | --- |
+| `tsc --noEmit` | 0 错 |
+| `npm run lint` | 0 error（既有 warning 未动） |
+| `npm test` | 4197（基线 4192，+5）；pass 4143 / fail 24 / skipped 30；24 条失败按名称与基线逐条一致（无 Docker 的 DB / 部署套件） |
+| `next build --webpack` | exit 0；`/`、`/chart`、`/ephemeris`、`/people` 仍 ○ Static；首屏 gzip 以执行方实测 130,933 B（=基线）为准，补修只改服务端与 lib 类型 |
+| 真机 | 未做（无登录态、无模型凭据），见 `docs/testing/consult-answer-the-question-20260927.md` 8 条 |
+
+结论：T1–T6 通过；补修一处（寒暄回复误判追问）已并入本分支。推 staging 后核对 `/api/health` 的 `deployment.gitCommit`。

docs/testing/consult-answer-the-question-20260927.md

@@ -13,5 +13,6 @@
 | 5 | 第 1～4 步过程中看步骤栏 | 第一节标题是「先回答你问的这件事」；首轮它的步骤是「用盘上的反差回答这个问题」，第 2、3 步是「直接回答这句话」；没有「先抓住你真正在问的事」；写作行不出现「正在写先回答你的问题」 | |
 | 6 | 第 2 步的回答完成后，看用量 / 后台 | 仍然计一次排盘、扣一次点（追问轮照常调用排盘工具） | |
 | 7 | 换无出生分钟的星盘问「今天适合做什么」 | 开场结尾仍有一到三条短行动（无骨架模式行动留在开场）；不出现本命结构 | |
+| 8 | 新会话先发「你好」，收到一句寒暄回复后再问「我和父母的关系怎么看」 | 仍是第 1 步的首轮形状（口语开场 + 三个二级标题），不是 200 字追问形状（寒暄回复不算上一轮解读） | |
 
 任一条不符，按 BUG-1070～1073 追加记录，附问句句式与回答的**结构描述**（标题、字数、第一句是不是是 / 不是），不贴回答原文。

frontend/src/app/api/consult/route.ts

@@ -995,8 +995,8 @@ export async function POST(request: Request) {
     // A session that already holds an answer gets the follow-up shape
     // (BUG-1071): a direct answer, no skeleton. That is read from the stored
     // history, never from the wording of the question.
-    const followUpTurn = hasPriorAssistantAnswer(history);
-    const natalInstruction = `${natalToolInstruction}${natalUserTurnShape({ entrypoint: consultEntrypoint, history })}`;
+    const followUpTurn = hasPriorAssistantAnswer(history, { summaryText: historyWindow.summaryText });
+    const natalInstruction = `${natalToolInstruction}${natalUserTurnShape({ entrypoint: consultEntrypoint, history, summaryText: historyWindow.summaryText })}`;
     const adoptedRangeNote = consultationMode === "verified_chart"
       && prepared.serverChart?.truth.birthTimeStatus === "accepted"
       && prepared.serverChart.toolInput.candidate_range
@@ -1518,7 +1518,7 @@ export async function POST(request: Request) {
     const workflowReceipt = consultationWorkflowReceipt(workflowContext);
     // The grounded fallback writes from the injected workflow; same first-turn
     // / follow-up split as the agentic path (BUG-1071), one statement of it.
-    const legacyNatalInstruction = hasPriorAssistantAnswer(history)
+    const legacyNatalInstruction = hasPriorAssistantAnswer(history, { summaryText: historyWindow.summaryText })
       ? followUpAnswerShapeInstruction()
       : "先用不超过 400 字口语直接回答下面的问题，不要加标题；形状为：先给反差（表面 A，底下 B，都是盘上的结构，命名成一个格局）、再说清谁在推谁在修、然后「别去应 X 的象，去扮演 Y 的象」；开场不写行动。然后按三个二级标题写：## 盘里支持这个判断的地方（原始结构、六步宫位、Yoga 表作为依据，只写开场没说过的，表格与正文二选一）、## 时间怎么看、## 这周可以做的一件事（最多三条行动，只在这里出现一次）。全文不超过 900 字。骨架不可省略；不写「先回答你的问题」这个标题，文末技法审计表由产品界面折叠展示，不写进正文。星盘事实只使用系统里已经注入的计算结果，不要复述内部字段、JSON 或再跑一遍咨询流程。";
 

frontend/src/lib/consultation-session-history.ts

@@ -15,6 +15,8 @@ export const SESSION_CONTEXT_SUMMARY_HEADING = "【会话摘要（服务端维
 export type ConsultationHistoryMessage = Readonly<{
   role: "user" | "assistant";
   text: string;
+  /** A stored smalltalk reply (BUG-976/977 fast path); it is not a natal answer. */
+  responseKind?: "smalltalk";
 }>;
 
 export type SessionContextSummaryV1 = Readonly<{
@@ -37,6 +39,7 @@ type StoredTurn = Readonly<{
   role: "user" | "assistant";
   text: string;
   requestId: string | null;
+  responseKind?: "smalltalk";
 }>;
 
 function clamp(value: number, min: number, max: number): number {
@@ -107,7 +110,7 @@ export function storedConsultationTurns(
   const rows: StoredTurn[] = [];
   messages.forEach((message, index) => {
     if (!message || typeof message !== "object") return;
-    const stored = message as { role?: unknown; text?: unknown; requestId?: unknown };
+    const stored = message as { role?: unknown; text?: unknown; requestId?: unknown; responseKind?: unknown };
     if (options.excludeRequestId && stored.requestId === options.excludeRequestId) return;
     if (stored.role !== "user" && stored.role !== "assistant") return;
     if (typeof stored.text !== "string" || !stored.text) return;
@@ -116,6 +119,7 @@ export function storedConsultationTurns(
       role: stored.role,
       text: stored.text,
       requestId: typeof stored.requestId === "string" ? stored.requestId : null,
+      ...(stored.responseKind === "smalltalk" ? { responseKind: "smalltalk" as const } : {}),
     });
   });
   return rows;
@@ -156,6 +160,7 @@ export function consultationHistoryWindow(
   const clipped = selected.map((turn) => ({
     role: turn.role,
     text: clipConsultationHistoryText(turn.text),
+    ...(turn.responseKind ? { responseKind: turn.responseKind } : {}),
   }));
   const budget = historyBudgetChars(options.contextWindow);
   let droppedCount = 0;

frontend/src/lib/consultation-thinking-plan.ts

@@ -383,11 +383,21 @@ export function followUpAnswerShapeInstruction(): string {
   ].join("");
 }
 
-/** The stored history already holds an assistant answer: this turn is a follow-up (D6, no intent regex). */
+/**
+ * The session already holds a natal answer, so this turn is a follow-up (D6,
+ * no intent regex). A stored smalltalk reply (「你好」 fast path) is an
+ * assistant message but not an answer, so it does not count; a session
+ * context summary only exists after earlier consultation turns were
+ * summarized away, so it counts even when the tail no longer holds them.
+ */
 export function hasPriorAssistantAnswer(
-  history: ReadonlyArray<Readonly<{ role: "user" | "assistant"; text: string }>>,
+  history: ReadonlyArray<Readonly<{ role: "user" | "assistant"; text: string; responseKind?: "smalltalk" }>>,
+  options: Readonly<{ summaryText?: string | null }> = {},
 ): boolean {
-  return history.some((message) => message.role === "assistant" && message.text.trim().length > 0);
+  if (options.summaryText?.trim()) return true;
+  return history.some((message) => (
+    message.role === "assistant" && message.responseKind !== "smalltalk" && message.text.trim().length > 0
+  ));
 }
 
 /**
@@ -397,10 +407,11 @@ export function hasPriorAssistantAnswer(
  */
 export function natalUserTurnShape(input: {
   entrypoint?: string;
-  history: ReadonlyArray<Readonly<{ role: "user" | "assistant"; text: string }>>;
+  history: ReadonlyArray<Readonly<{ role: "user" | "assistant"; text: string; responseKind?: "smalltalk" }>>;
+  summaryText?: string | null;
 }): string {
   if (input.entrypoint === "daily_starlanguage") return dailyAnswerShapeInstruction();
-  return hasPriorAssistantAnswer(input.history)
+  return hasPriorAssistantAnswer(input.history, { summaryText: input.summaryText })
     ? followUpAnswerShapeInstruction()
     : natalAnswerShapeInstruction();
 }

frontend/src/mastra/product-voice.ts

@@ -66,6 +66,6 @@ Raman yogas and house steps remain governed_support_only_candidate_ready: write
 Never:
 - Skip a card fact the question needs in order to keep the answer short — and never write a card fact the question does not need in order to look thorough.
 - Put the Technique Audit Table or 统一参数 heading back into the spoken body.
-- State what the user or the person asked about wants, needs, or feels; state where the chart puts them.
+- State what the user or the person asked about wants, needs, or feels. Say where the chart puts them instead.
 
 The first natal career/wealth/marriage/family answer of a session uses the opener-plus-skeleton above; a follow-up turn follows the shape instruction carried in the user turn (a direct answer of about 200 characters, no headings) unless the user changes domain or asks for the full reading.`;

frontend/tests/consult-single-pass-answer-20260927.test.ts

@@ -432,8 +432,11 @@ test("the consult route has no separate compose stream and wires the single-pass
   // 原值: assert.match(route, /natalAnswerShapeInstruction\(\)/)
   // 新值: 用户轮形状经 natalUserTurnShape 按「会话已有回答」切换首轮 / 追问轮，思考栏与工具上下文带 followUpTurn
   // 原因: TASK-consult-answer-the-question-20260927 D3/D6（BUG-1071）
-  assert.match(route, /natalUserTurnShape\(\{ entrypoint: consultEntrypoint, history \}\)/);
-  assert.match(route, /const followUpTurn = hasPriorAssistantAnswer\(history\);/);
+  // 原值: natalUserTurnShape({ entrypoint, history }) / hasPriorAssistantAnswer(history)
+  // 新值: 两处都带 summaryText（Claude 验收补修：寒暄回复不算上一轮解读，摘要算）
+  // 原因: 「你好」的回复也存成 assistant 消息，只看 role 会把第一个正式问题当追问
+  assert.match(route, /natalUserTurnShape\(\{ entrypoint: consultEntrypoint, history, summaryText: historyWindow\.summaryText \}\)/);
+  assert.match(route, /const followUpTurn = hasPriorAssistantAnswer\(history, \{ summaryText: historyWindow\.summaryText \}\);/);
   assert.match(route, /followUp: followUpTurn,/);
   assert.match(route, /followUpTurn,\n\s+serverChart: prepared\.serverChart,/);
   assert.doesNotMatch(route, /natalAnswerShapeInstruction\(\)/);

frontend/tests/consultation-session-history.test.ts

@@ -73,6 +73,19 @@ test("a 20_000 character message is clipped to 12_000 with an omission count", (
   assert.equal(history.tail[0]?.text, clipped);
 });
 
+test("a stored smalltalk reply keeps its responseKind through the history window", () => {
+  // Claude acceptance fix for the follow-up detection: the route reads responseKind
+  // off the window tail to tell a 「你好」 reply from a natal answer.
+  const history = consultationHistoryWindow([
+    { role: "user", text: "你好" },
+    { role: "assistant", text: "你好，想聊什么都可以", responseKind: "smalltalk" },
+    { role: "user", text: "我和父母的亲疏关系如何" },
+    { role: "assistant", text: "外顺内远。" },
+  ], null, { contextWindow: 128_000 });
+  assert.equal(history.tail[1]?.responseKind, "smalltalk");
+  assert.equal("responseKind" in (history.tail[3] ?? {}), false);
+});
+
 test("historyBudgetChars uses 64k, 32k, and null-as-128k windows", () => {
   assert.equal(historyBudgetChars(64_000), 6_000);
   assert.equal(historyBudgetChars(32_000), 4_000);

frontend/tests/consultation-thinking-plan.test.ts

@@ -232,6 +232,20 @@ test("a follow-up turn answers directly and the thinking bar says so (TASK-consu
     { role: "user", text: "我和父母的亲疏关系如何" },
     { role: "assistant", text: "外顺内远。" },
   ]), true);
+  // Claude acceptance fix: a stored smalltalk reply (「你好」 fast path, responseKind "smalltalk")
+  // is an assistant message but not a natal answer, so the first real question after it
+  // still gets the first-turn shape; a session summary means earlier answers were folded away.
+  assert.equal(hasPriorAssistantAnswer([
+    { role: "user", text: "你好" },
+    { role: "assistant", text: "你好，想聊什么都可以", responseKind: "smalltalk" },
+  ]), false);
+  assert.equal(hasPriorAssistantAnswer([], { summaryText: "此前一轮讲过父母：外顺内远。" }), true);
+  assert.equal(hasPriorAssistantAnswer([], { summaryText: "   " }), false);
+  assert.equal(
+    natalUserTurnShape({ history: [{ role: "user", text: "你好" }, { role: "assistant", text: "你好", responseKind: "smalltalk" }] }),
+    natalAnswerShapeInstruction(),
+  );
+  assert.equal(natalUserTurnShape({ history: [], summaryText: "此前一轮讲过父母。" }), followUpAnswerShapeInstruction());
 
   // ① empty history → first-turn shape; ② prior answer → follow-up; ④ daily is untouched by history.
   const first = natalUserTurnShape({ history: [] });

frontend/tests/consultation-voice-contract.test.ts

@@ -182,7 +182,9 @@ test("the voice answers the sentence asked: no motive guessing, yes/no first, co
   assert.match(voice, /Good \(是非题，用户问「我妈是不是不太在意我」/);
   assert.match(voice, /不是。4 宫主水星逆行落 10 宫/);
   assert.match(voice, /Bad \(同一问题\): 你其实要的是被听见/);
-  assert.match(voice, /State what the user or the person asked about wants, needs, or feels; state where the chart puts them/);
+  // 原值: "…wants, needs, or feels; state where the chart puts them"（Never 列表里第二句读成也被禁止）
+  // 新值: 两句拆开，第二句是替代做法  原因: Claude 验收改措辞
+  assert.match(voice, /State what the user or the person asked about wants, needs, or feels\. Say where the chart puts them instead/);
   // The three-mode shared voice still carries all of it (general and window expand the same literal).
   const sharedVoice = voice.slice(0, voice.indexOf("export const natalSpokenReportContract"));
   assert.ok(sharedVoice.includes("不揣测（回答对象是用户问的那句话）"));