fix(web): stop consultation thinking from pinching the answer

Disable provider thinking so Flash CoT cannot fill max_tokens, raise the spoken budget to 16384, emit a server-owned step tree, and continue once when the body ends on length. Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-22 21:19:00 +08:00
parent d2a5f0f5bf
commit 0ca7da997f
26 changed files with 1013 additions and 88 deletions
@@ -36,6 +36,7 @@ import { createServerSupabaseClient } from "@/lib/supabase/server";
 import { streamTextResponse } from "@/lib/stream-text-response";
 import { streamAgentResponse } from "@/lib/stream-agent-response";
 import type { AgentExecutionReceipt, WorkflowReceipt } from "@/lib/consultation-agent-events";
+import { consultationContinuePrompt, type PublicThinkingSection } from "@/lib/consultation-thinking-plan";
 import {
  AGENT_MAX_STEPS,
  AGENT_TIMEOUT_MS,
@@ -501,6 +502,7 @@ export async function POST(request: Request) {
    workflowReceipt: WorkflowReceipt,
    agentExecutionReceipt?: AgentExecutionReceipt,
    thinkingText?: string,
+    thinkingSections?: PublicThinkingSection[],
  ): Promise<AgentSettlementResult> {
    try {
      const reply = parseAgentReply(
@@ -513,6 +515,7 @@ export async function POST(request: Request) {
        role: "assistant" as const,
        text: reply.text,
        ...(persistedThinking ? { thinkingText: persistedThinking } : {}),
+        ...(thinkingSections?.length ? { thinkingSections } : {}),
        techniqueTruth,
        workflowReceipt,
        ...(agentExecutionReceipt ? { agentExecutionReceipt } : {}),
@@ -724,6 +727,15 @@ export async function POST(request: Request) {
        usages.push(retried.totalUsage);
        return retried.fullStream;
      };
+      const continueAfterLength = async (output: string) => {
+        const continued = await agent.stream([
+          ...baseMessages,
+          { role: "assistant" as const, content: output },
+          { role: "user" as const, content: consultationContinuePrompt(output) },
+        ], streamOptions);
+        usages.push(continued.totalUsage);
+        return continued.fullStream;
+      };
      const executionReceipt = (): AgentExecutionReceipt => ({
        runId: requestId,
        runtime: "mastra-agentic",
@@ -748,6 +760,7 @@ export async function POST(request: Request) {
        stream: result.fullStream,
        requireTool: false,
        retryForAnswer,
+        continueAfterLength,
        continueAfterDisconnect: true,
        transformText: createBirthTimeModeOutputGuard(
          generalDailyContext ? "general_no_birth_time" : consultationMode,
@@ -758,13 +771,14 @@ export async function POST(request: Request) {
        headers: { "x-jyotish-birth-time-mode": consultationMode },
        onFirstActivity: markFirstActivity,
        onFirstOutput: markFirstText,
-        onComplete: (output, agentExecutionReceipt, thinkingText) => settleRun(() => completeResponse(
+        onComplete: (output, agentExecutionReceipt, thinkingText, thinkingSections) => settleRun(() => completeResponse(
          output,
          mergeUsage(usages),
          generalDailyContext ? "public-panchanga-only" : "not-applicable",
          workflowReceipt,
          agentExecutionReceipt,
          thinkingText,
+          thinkingSections,
        ), undefined),
        onError: (error) => settleRun(
          cancel,
@@ -812,6 +826,15 @@ export async function POST(request: Request) {
        usages.push(retried.totalUsage);
        return retried.fullStream;
      };
+      const continueAfterLength = async (output: string) => {
+        const continued = await agent.stream([
+          ...baseMessages,
+          { role: "assistant" as const, content: output },
+          { role: "user" as const, content: consultationContinuePrompt(output) },
+        ], streamOptions);
+        usages.push(continued.totalUsage);
+        return continued.fullStream;
+      };
      const executionReceipt = (): AgentExecutionReceipt => ({
        runId: requestId,
        runtime: "mastra-agentic",
@@ -837,6 +860,7 @@ export async function POST(request: Request) {
        requireTool: true,
        retry,
        retryForAnswer,
+        continueAfterLength,
        continueAfterDisconnect: true,
        transformText: createBirthTimeModeOutputGuard(consultationMode, false),
        toolStatus: () => workflowStatus(state.workflowReceipt?.status),
@@ -844,13 +868,14 @@ export async function POST(request: Request) {
        headers: { "x-jyotish-birth-time-mode": consultationMode },
        onFirstActivity: markFirstActivity,
        onFirstOutput: markFirstText,
-        onComplete: (output, agentExecutionReceipt, thinkingText) => settleRun(() => completeResponse(
+        onComplete: (output, agentExecutionReceipt, thinkingText, thinkingSections) => settleRun(() => completeResponse(
          output,
          mergeUsage(usages),
          state.techniqueTruth ?? "declared-window",
          state.workflowReceipt ?? workflowReceipt,
          agentExecutionReceipt,
          thinkingText,
+          thinkingSections,
        ), undefined),
        onError: (error) => settleRun(
          cancel,
@@ -900,6 +925,15 @@ export async function POST(request: Request) {
      usages.push(retried.totalUsage);
      return retried.fullStream;
    };
+    const continueAfterLength = async (output: string) => {
+      const continued = await agent.stream([
+        ...baseMessages,
+        { role: "assistant" as const, content: output },
+        { role: "user" as const, content: consultationContinuePrompt(output) },
+      ], streamOptions);
+      usages.push(continued.totalUsage);
+      return continued.fullStream;
+    };
    const executionReceipt = (): AgentExecutionReceipt => ({
      runId: requestId,
      runtime: "mastra-agentic",
@@ -925,6 +959,7 @@ export async function POST(request: Request) {
      requireTool: true,
      retry,
      retryForAnswer,
+      continueAfterLength,
      continueAfterDisconnect: true,
      transformText: (text) => createBirthTimeModeOutputGuard(
        consultationMode,
@@ -935,13 +970,14 @@ export async function POST(request: Request) {
      headers: { "x-jyotish-birth-time-mode": consultationMode },
      onFirstActivity: markFirstActivity,
      onFirstOutput: markFirstText,
-      onComplete: (output, agentExecutionReceipt, thinkingText) => settleRun(() => completeResponse(
+      onComplete: (output, agentExecutionReceipt, thinkingText, thinkingSections) => settleRun(() => completeResponse(
        output,
        mergeUsage(usages),
        state.techniqueTruth ?? "unknown",
        state.workflowReceipt ?? workflowReceipt,
        agentExecutionReceipt,
        thinkingText,
+        thinkingSections,
      ), undefined),
      onError: (error) => settleRun(
        cancel,
@@ -844,6 +844,37 @@ button:disabled { cursor: default; opacity: .45; }
  line-height: 1.55;
  white-space: pre-wrap;
 }
+.consultation-step-tree__intent {
+  margin: 0 0 var(--space-2);
+  color: var(--color-ink-secondary);
+  font-size: 13px;
+  line-height: 1.5;
+}
+.agent-thinking-marker.is-pending {
+  background: var(--color-canvas-muted);
+  box-shadow: inset 0 0 0 1px var(--color-border);
+}
+.agent-thinking-step.is-more {
+  color: var(--color-ink-tertiary);
+}
+.consultation-thinking-report {
+  display: grid;
+  gap: var(--space-5);
+}
+.consultation-report-block {
+  display: grid;
+  gap: var(--space-2);
+}
+.consultation-report-analysis__label {
+  margin: 0 0 var(--space-2);
+  color: var(--color-ink);
+  font-size: 13px;
+  font-weight: 600;
+  line-height: 1.5;
+}
+.consultation-report-analysis .message-answer {
+  margin-top: 0;
+}
 .rectification-message-entry { min-width: 0; }
 .message-actions {
  display: flex;
@@ -106,6 +106,12 @@ import {
  type AgentExecutionReceipt,
  type ConsultationAgentPublicEvent,
 } from "@/lib/consultation-agent-events";
+import {
+  applyThinkingSectionProgress,
+  parsePublicThinkingSections,
+  upsertThinkingSection,
+  type PublicThinkingSection,
+} from "@/lib/consultation-thinking-plan";
 import {
  CONSULTATION_CHART_CALCULATION_LABEL,
  CONSULTATION_COMPOSING_LABEL,
@@ -220,7 +226,13 @@ type ReplyOutcome = {
  readonly phase: Extract<ChatReplyPhase, "completed" | "stopped" | "failed">;
  readonly replyOrdinal: number;
 };
-type StreamingReply = { sessionId: string; text: string; activity?: AgentActivityView; thinkingText?: string };
+type StreamingReply = {
+  sessionId: string;
+  text: string;
+  activity?: AgentActivityView;
+  thinkingText?: string;
+  thinkingSections?: PublicThinkingSection[];
+};
 type BirthPlace = {
  label: string;
  lat: number;
@@ -780,10 +792,12 @@ function readSessions(value: unknown, catalog: PublicLanguageModelCatalog | null
        const thinkingText = typeof stored.thinkingText === "string" && stored.thinkingText.trim()
          ? stored.thinkingText.slice(0, 4000)
          : undefined;
+        const thinkingSections = parsePublicThinkingSections(stored.thinkingSections);
        return [{
          role: stored.role,
          text: stored.text.slice(0, 12000),
          ...(thinkingText ? { thinkingText } : {}),
+          ...(thinkingSections.length ? { thinkingSections } : {}),
          ...(typeof stored.techniqueTruth === "string" ? { techniqueTruth: stored.techniqueTruth } : {}),
          ...(stored.agentExecutionReceipt ? { agentExecutionReceipt: stored.agentExecutionReceipt } : {}),
          ...(stored.workflowReceipt ? { workflowReceipt: stored.workflowReceipt } : {}),
@@ -1197,6 +1211,9 @@ export default function Home() {
  const activeStreamingThinking = streamingReply && streamingReply.sessionId === activeSession?.id
    ? streamingReply.thinkingText
    : undefined;
+  const activeStreamingSections = streamingReply && streamingReply.sessionId === activeSession?.id
+    ? streamingReply.thinkingSections
+    : undefined;
  const activeReplyOutcome = replyOutcome && replyOutcome.sessionId === activeSession?.id ? replyOutcome : null;
  const replyPhase: ChatReplyPhase = isLoading
    ? consultationPhase === "recovering" ? "recovering" : "generating"
@@ -2015,6 +2032,7 @@ export default function Home() {
        role: message.role,
        text: message.text,
        thinkingText: message.thinkingText,
+        thinkingSections: message.thinkingSections,
        techniqueTruth: message.techniqueTruth,
        agentExecutionReceipt: message.agentExecutionReceipt,
        workflowReceipt: message.workflowReceipt,
@@ -3249,7 +3267,7 @@ export default function Home() {
    }
    setStreamingReply({ sessionId, text: "" });
    let latestPartialReply = "";
-    let thinking = "";
+    let thinkingSections: PublicThinkingSection[] = [];
    try {
      const response = await fetch("/api/consult", {
        method: "POST",
@@ -3314,10 +3332,11 @@ export default function Home() {
      const updateStreamingAnswer = (activity?: AgentActivityView) => {
        const partialReply = parseAgentReply(answer).text;
        latestPartialReply = partialReply;
+        thinkingSections = applyThinkingSectionProgress(thinkingSections, partialReply);
        setStreamingReply((current) => ({
          sessionId,
          text: partialReply,
-          thinkingText: current?.sessionId === sessionId ? current.thinkingText : undefined,
+          thinkingSections: thinkingSections.length ? thinkingSections : undefined,
          activity: activity
            ? nextActivityView(current?.sessionId === sessionId ? current.activity : undefined, activity)
            : current?.sessionId === sessionId ? current.activity : undefined,
@@ -3359,12 +3378,20 @@ export default function Home() {
      if ((response.headers.get("content-type") ?? "").includes("application/x-ndjson")) {
        const parser = createNdjsonParser((event) => {
          if (event.type === "answer.delta") answer += event.text;
-          if (event.type === "thinking.delta") {
-            thinking = `${thinking}${event.text}`.slice(0, 4_000);
+          if (event.type === "thinking.section") {
+            thinkingSections = applyThinkingSectionProgress(
+              upsertThinkingSection(thinkingSections, {
+                id: event.id,
+                title: event.title,
+                heading: event.heading,
+                steps: event.steps,
+              }),
+              parseAgentReply(answer).text,
+            );
            setStreamingReply((current) => ({
              sessionId,
              text: current?.sessionId === sessionId ? current.text : parseAgentReply(answer).text,
-              thinkingText: thinking,
+              thinkingSections,
              activity: current?.sessionId === sessionId ? current.activity : undefined,
            }));
          }
@@ -3403,7 +3430,7 @@ export default function Home() {
            messages: [...userSession.messages, {
              role: "assistant",
              text: reply.text,
-              thinkingText: thinking || undefined,
+              ...(thinkingSections.length ? { thinkingSections } : {}),
              techniqueTruth,
              workflowReceipt,
              agentExecutionReceipt,
@@ -3429,8 +3456,8 @@ export default function Home() {
        if (!runCompleted && !truncatedFailure) {
          throw new ConsultationResponseError(
            502,
-            thinking.trim()
-              ? "这次还没有生成可显示的回答。思考过程已保留，可以直接继续问。"
+            thinkingSections.length
+              ? "这次还没有生成可显示的回答。思考步骤已保留，可以直接继续问。"
              : "Agent 回答未完成，本次不会保存为成功咨询。",
          );
        }
@@ -3446,8 +3473,8 @@ export default function Home() {
      if (controller.signal.aborted) return Boolean(latestPartialReply);
      const reply = parseAgentReply(answer);
      if (!reply.text) {
-        throw thinking.trim()
-          ? new ConsultationResponseError(502, "这次还没有生成可显示的回答。思考过程已保留，可以直接继续问。")
+        throw thinkingSections.length
+          ? new ConsultationResponseError(502, "这次还没有生成可显示的回答。思考步骤已保留，可以直接继续问。")
          : new Error("Agent 没有返回可显示的回答，请重试。");
      }

@@ -3464,7 +3491,7 @@ export default function Home() {
        messages: [...userSession.messages, {
          role: "assistant",
          text: reply.text,
-          thinkingText: thinking || undefined,
+          ...(thinkingSections.length ? { thinkingSections } : {}),
          techniqueTruth,
          workflowReceipt,
          agentExecutionReceipt,
@@ -3508,13 +3535,13 @@ export default function Home() {
        if (restore) {
          updateSession(sessionId, () => restore);
          void persistSession(restore).catch(() => {});
-        } else if (thinking.trim() || latestPartialReply) {
+        } else if (thinkingSections.length || latestPartialReply) {
          const failedSession: ChatSession = {
            ...userSession,
            messages: [...userSession.messages, {
              role: "assistant",
              text: latestPartialReply,
-              thinkingText: thinking.trim() || undefined,
+              ...(thinkingSections.length ? { thinkingSections } : {}),
            }],
            updatedAt: timestamp(),
          };
@@ -3950,7 +3977,7 @@ export default function Home() {
            </div>
          ) : (
            <div className="message-list" aria-busy={isLoading}>
-              {chatMessageViews(activeSession.messages, isLoading, activeStreamingText, activeStreamingActivity, activeStreamingThinking).map((message, index, views) => {
+              {chatMessageViews(activeSession.messages, isLoading, activeStreamingText, activeStreamingActivity, activeStreamingThinking, activeStreamingSections).map((message, index, views) => {
                const showActions = message.role === "assistant"
                  && message.state === "settled"
                  && Boolean(message.text);