feat(consult): open the response stream before preparation, reservation and classification; rejections ride the stream (T2, BUG-1074)

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

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

@@ -40,6 +40,7 @@ import { createServerSupabaseClient } from "@/lib/supabase/server";
 import { streamTextResponse } from "@/lib/stream-text-response";
 import { classifyConsultationTurn, type SmalltalkUsage } from "@/lib/consultation-smalltalk";
 import { streamSmalltalkResponse } from "@/lib/stream-smalltalk-response";
+import { streamFirstResponse } from "@/lib/stream-first-response";
 import { consultationPublicActivityEvent, streamAgentResponse } from "@/lib/stream-agent-response";
 import type { AgentExecutionReceipt, WorkflowReceipt } from "@/lib/consultation-agent-events";
 import {
@@ -283,6 +284,7 @@ function workflowStatus(status: string | undefined): "ready" | "degraded" | "blo
 }
 
 export async function POST(request: Request) {
+  const requestStartedAt = Date.now();
   let supabase: Awaited<ReturnType<typeof createServerSupabaseClient>>;
   let accounting: ReturnType<typeof createAdminSupabaseClient>;
   try {
@@ -406,6 +408,16 @@ export async function POST(request: Request) {
   type SelectedModel = NonNullable<Awaited<ReturnType<typeof resolveSessionLanguageModel>>>;
   type ReservationResult = { success: boolean; credits: number | null; error_code: string | null };
   type ModelSelection = Awaited<ReturnType<typeof reserveConsultationModel<SelectedModel, ReservationResult>>>;
+
+  // Everything above answered with a real HTTP status: authentication, the
+  // request body, the session and its model. Everything below — subject and
+  // profile reads, the credit reservation, appending the question, the model
+  // classification, the agent itself — runs inside the response stream in the
+  // agentic runtime (BUG-1074): `runTurn` still returns the same Responses it
+  // always did, and `streamFirstResponse` relays them after a first frame that
+  // goes out before any of this work starts. The legacy runtime keeps the
+  // pre-stream order because its text/plain replies cannot ride an NDJSON stream.
+  const runTurn = async (): Promise<Response> => {
   let prepared: PreparedConsultationRoute<ModelSelection, ConsultationPlan>;
   try {
     prepared = await prepareConsultationRoute<ModelSelection, ConsultationPlan>({
@@ -1625,4 +1637,25 @@ export async function POST(request: Request) {
       { status: 503 },
     );
   }
+  };
+
+  if (!shouldUseAgenticRuntime(user)) return runTurn();
+  return streamFirstResponse({
+    requestId,
+    run: runTurn,
+    startedAt: requestStartedAt,
+    onFirstFrame: (elapsedMs) => {
+      logAgentObservability({
+        requestId,
+        sessionId,
+        agentVersion: "consultation-stream-first-v1",
+        contractPhases: [{ phase: "first_byte", durationMs: elapsedMs, status: "completed" }],
+        billingSettlementResult: "not_applicable",
+      });
+    },
+    onUnhandled: (error) => {
+      const reason = error instanceof Error ? error.name : "UnknownError";
+      console.error(`[consult] turn failed before a response request=${requestId} reason=${reason}`);
+    },
+  });
 }