feat(consult): type out streamed text at a capped pace and write the tail out on settle instead of one frame (T3, BUG-1075)

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

frontend/src/hooks/use-consultation-run.ts

@@ -489,7 +489,9 @@ export function useConsultationRun(params: ConsultationRunParams) {
     if (pendingConsultation.current?.requestId !== requestId) return;
     pendingConsultation.current = null;
     if (consultationReplayStarted.current === requestId) consultationReplayStarted.current = null;
-    setStreamingReply(null);
+    // A reply still being written out at the typing pace keeps the display
+    // (BUG-1075); its last paced frame clears it. Everything else is done now.
+    setStreamingReply((current) => (current?.settling ? current : null));
     setPendingSessionId(null);
     setPendingRequestId(null);
     setConsultationPhase(null);
@@ -766,6 +768,11 @@ export function useConsultationRun(params: ConsultationRunParams) {
     let streamedThinking = "";
     let timelineState = emptyConsultationTimeline();
     let currentActivity: AgentActivityView | undefined;
+    // True once the stream has ended and the buffer is writing out what is
+    // left at the typing pace (BUG-1075). From then on the display belongs to
+    // this request only while the streaming reply still carries its id: a new
+    // send replaces the reply, and the old buffer's frames fall through.
+    let settleRequested = false;
     // Every stream event lands in `frames`; it commits at most once per animation
     // frame and releases text at a steady pace. Nothing below calls
     // setStreamingReply directly while the response body is being read.
@@ -775,7 +782,7 @@ export function useConsultationRun(params: ConsultationRunParams) {
         const partialReply = parseAgentReply(frame.answer).text;
         latestPartialReply = partialReply;
         thinkingSections = applyThinkingSectionProgress(thinkingSections, partialReply);
-        setStreamingReply({
+        const next: StreamingReply = {
           sessionId,
           text: partialReply,
           responseKind,
@@ -783,12 +790,33 @@ export function useConsultationRun(params: ConsultationRunParams) {
           thinkingSections: thinkingSections.length ? thinkingSections : undefined,
           timeline: timelineState.rows,
           activity: currentActivity,
-        });
+          ...(settleRequested ? { settling: requestId } : {}),
+        };
+        if (settleRequested) {
+          setStreamingReply((current) => (
+            current?.settling !== requestId ? current : frame.settled ? null : next
+          ));
+        } else {
+          setStreamingReply(next);
+        }
         if (partialReply && pendingConsultation.current?.requestId === requestId) {
           pendingConsultation.current = { ...pendingConsultation.current, partialReply };
         }
       },
     });
+    // The stream has ended: write out the rest at the typing pace unless the
+    // run was stopped or failed, in which case what arrived shows at once.
+    const settleFrames = (immediate: boolean) => {
+      if (immediate) {
+        void frames.settle({ immediate: true });
+        return;
+      }
+      settleRequested = true;
+      setStreamingReply((current) => (
+        current && current.sessionId === sessionId ? { ...current, settling: requestId } : current
+      ));
+      void frames.settle();
+    };
     try {
       const response = await fetch("/api/consult", {
         method: "POST",
@@ -949,7 +977,7 @@ export function useConsultationRun(params: ConsultationRunParams) {
           parser.push(decoder.decode(value, { stream: true }));
         }
         parser.finish(decoder.decode());
-        frames.settle();
+        settleFrames(controller.signal.aborted || Boolean(truncatedFailure));
         if (truncatedFailure) {
           const reply = parseAgentReply(answer);
           if (!reply.text) throw new ConsultationResponseError(502, truncatedFailure.message);
@@ -992,7 +1020,7 @@ export function useConsultationRun(params: ConsultationRunParams) {
         }
         answer += decoder.decode();
         frames.setAnswer(answer);
-        frames.settle();
+        settleFrames(controller.signal.aborted);
       }
       if (controller.signal.aborted) return Boolean(latestPartialReply);
       const reply = parseAgentReply(answer);
@@ -1043,7 +1071,8 @@ export function useConsultationRun(params: ConsultationRunParams) {
     } catch (caught) {
       // Whatever arrived before the failure is what gets kept, not just the
       // part the pacing had released so far.
-      frames.settle();
+      settleRequested = false;
+      void frames.settle({ immediate: true });
       const cancelled = controller.signal.aborted;
       const ownsInterface = pendingConsultation.current?.requestId === requestId;
       const partialReply = latestPartialReply;