fix(onboard): honor back at the NVIDIA API key prompt

src/lib/onboard.ts

@@ -26,6 +26,7 @@ const {
   createNvidiaFeaturedModelSession,
   createRemoteModelValidator,
   resolveCompatibleEndpointSelection,
+  selectFeaturedModelAfterCredentialPrompt,
 }: typeof import("./onboard/setup-nim-selection") = require("./onboard/setup-nim-selection");
 const setupNimFlow: typeof import("./onboard/setup-nim-flow") = require("./onboard/setup-nim-flow");
 const openrouterSelection: typeof import("./onboard/openrouter-selection") = require("./onboard/openrouter-selection");
@@ -2539,6 +2540,7 @@ async function handleRemoteProviderSelection(args: RemoteProviderSelectionArgs,
   hydrateCredentialEnv(state.credentialEnv);
   if (selected.key === "build") {
     providerKeyBridge.stageBuildProviderKeyBridge();
+    let apiKeyNavigation: unknown = null;
     if (isNonInteractive()) {
       const reuseGatewayCredential = buildCredentialReuse.resolveNonInteractiveBuildCredential({
         provider: state.provider,
@@ -2549,14 +2551,9 @@ async function handleRemoteProviderSelection(args: RemoteProviderSelectionArgs,
       state.skipHostInferenceSmoke = reuseGatewayCredential;
       state.reuseGatewayCredentialWithoutLocalKey = reuseGatewayCredential;
     } else {
-      await ensureApiKey();
+      apiKeyNavigation = await ensureApiKey();
     }
-    state.model = await state.nvidiaFeaturedModels!.select(
-      requestedModel || (typeof state.model === "string" ? state.model : null),
-      recoveredFromSandbox ? recoveredModel : null,
-      isNonInteractive(),
-      process.env.NEMOCLAW_MODEL,
-    );
+    state.model = await selectFeaturedModelAfterCredentialPrompt(state.nvidiaFeaturedModels!, apiKeyNavigation, credentialPrompt.shouldReturnToProviderSelection, requestedModel || (typeof state.model === "string" ? state.model : null), recoveredFromSandbox ? recoveredModel : null, isNonInteractive(), process.env.NEMOCLAW_MODEL);
     if (isBackToSelection(state.model)) {
       console.log("  Returning to provider selection.");
       console.log("");

src/lib/onboard/nvidia-featured-model-selection.test.ts

@@ -5,7 +5,12 @@ import { beforeEach, describe, expect, it, vi } from "vitest";
 
 import { promptCloudModel } from "../inference/model-prompts";
 import { BACK_TO_SELECTION } from "../navigation";
-import { createNvidiaFeaturedModelSession } from "./nvidia-featured-model-selection";
+import { shouldReturnToProviderSelection } from "./credential-navigation";
+import {
+  createNvidiaFeaturedModelSession,
+  type NvidiaFeaturedModelSession,
+  selectFeaturedModelAfterCredentialPrompt,
+} from "./nvidia-featured-model-selection";
 
 vi.mock("../inference/model-prompts", () => ({
   promptCloudModel: vi.fn(),
@@ -99,4 +104,51 @@ describe("NVIDIA featured model selection", () => {
       "environment/model",
     );
   });
+
+  it("skips the catalog when the NVIDIA API key prompt asks to go back (#9404)", async () => {
+    const select = vi.fn().mockResolvedValue("nvidia/selected-model");
+    const session = { select } as unknown as NvidiaFeaturedModelSession;
+    const exitOnboard = vi.fn(() => {
+      throw new Error("exit onboarding");
+    }) as unknown as () => never;
+    const shouldReturn = (result: unknown) => shouldReturnToProviderSelection(result, exitOnboard);
+
+    await expect(
+      selectFeaturedModelAfterCredentialPrompt(
+        session,
+        { kind: "back" },
+        shouldReturn,
+        null,
+        null,
+        false,
+      ),
+    ).resolves.toBe(BACK_TO_SELECTION);
+    expect(select).not.toHaveBeenCalled();
+    expect(exitOnboard).not.toHaveBeenCalled();
+
+    await expect(
+      selectFeaturedModelAfterCredentialPrompt(
+        session,
+        { kind: "credential", value: "nvapi-good" },
+        shouldReturn,
+        null,
+        null,
+        true,
+        "env/model",
+      ),
+    ).resolves.toBe("nvidia/selected-model");
+    expect(select).toHaveBeenCalledWith(null, null, true, "env/model");
+
+    await expect(
+      selectFeaturedModelAfterCredentialPrompt(
+        session,
+        { kind: "exit" },
+        shouldReturn,
+        null,
+        null,
+        false,
+      ),
+    ).rejects.toThrow("exit onboarding");
+    expect(select).toHaveBeenCalledTimes(1);
+  });
 });

src/lib/onboard/nvidia-featured-model-selection.ts

@@ -8,6 +8,7 @@ import {
   createNvidiaFeaturedModelPromptOptionsLoader,
   type NvidiaFeaturedModelOptions,
 } from "../inference/nvidia-featured-models";
+import { BACK_TO_SELECTION } from "../navigation";
 
 export type NvidiaFeaturedModelSession = {
   select: (
@@ -57,3 +58,20 @@ export function createNvidiaFeaturedModelSession(
     },
   };
 }
+
+/**
+ * Select a featured model only when the credential prompt did not ask to leave. `back` at the
+ * API key prompt must return to provider selection instead of loading the catalog (#9404).
+ */
+export async function selectFeaturedModelAfterCredentialPrompt(
+  session: NvidiaFeaturedModelSession,
+  credentialNavigation: unknown,
+  shouldReturnToProviderSelection: (result: unknown) => boolean,
+  requestedModel: string | null,
+  recoveredModel: string | null,
+  nonInteractive: boolean,
+  envModel?: string,
+): Promise<ModelPromptResult> {
+  if (shouldReturnToProviderSelection(credentialNavigation)) return BACK_TO_SELECTION;
+  return session.select(requestedModel, recoveredModel, nonInteractive, envModel);
+}

src/lib/onboard/setup-nim-selection.ts

@@ -15,7 +15,10 @@ import type { NvidiaFeaturedModelSession } from "./nvidia-featured-model-selecti
 import { exitOnboardFromPrompt, getNavigationChoice } from "./prompt-helpers";
 import type { ReasoningEffort } from "./reasoning-mode";
 
-export { createNvidiaFeaturedModelSession } from "./nvidia-featured-model-selection";
+export {
+  createNvidiaFeaturedModelSession,
+  selectFeaturedModelAfterCredentialPrompt,
+} from "./nvidia-featured-model-selection";
 
 export type SetupNimSelectionBackNavigation = Readonly<{ kind: "NEMOCLAW_BACK_TO_SELECTION" }>;