fix(onboard): explain unavailable providers

src/lib/onboard.ts

@@ -2577,16 +2577,6 @@ async function createSandboxWithBaseImageResolution(
     process.removeListener("exit", cleanupBuildCtx);
   }
 
-  if (manageDashboard) {
-    console.log("  Waiting for NemoClaw dashboard to become ready...");
-    sandboxReadinessTracing.waitForDashboardReadyWithTrace({
-      sandboxName,
-      port: effectiveDashboardPort,
-      runCaptureOpenshell,
-      sleep: sleepSeconds,
-    });
-  }
-
   if (effectiveSandboxGpuConfig.sandboxGpuEnabled) {
     await dockerGpuLocalInference.verifyGpuSandboxLocalInferenceAndCommitAfterReady(
       effectiveSandboxGpuConfig,

src/lib/onboard/provider-selection-failure.test.ts

@@ -15,6 +15,7 @@ function report(overrides: Partial<Parameters<typeof reportProviderSelectionFail
 
   reportProviderSelectionFailure({
     reason: { kind: "requested-provider-unavailable", providerKey: "missing" },
+    availableProviderKeys: ["build"],
     isWindowsHostOllama: false,
     rejectWindowsHostOllama: (providerKey, windowsHostSelected) => {
       rejected.push({ providerKey, windowsHostSelected });
@@ -85,14 +86,43 @@ describe("reportProviderSelectionFailure", () => {
     ]);
   });
 
-  it("reports unavailable requested providers", () => {
+  it("tells unavailable requested providers how to recover", () => {
     const { errors, rejected } = report({
       reason: { kind: "requested-provider-unavailable", providerKey: "missing-provider" },
     });
 
     assert.deepEqual(rejected, []);
     assert.deepEqual(errors, [
       "  Requested provider 'missing-provider' is not available in this environment.",
+      "  Re-run without NEMOCLAW_PROVIDER to choose an available provider, or install and start the requested provider before retrying.",
+    ]);
+  });
+
+  it("routes a missing vLLM server to the available managed runtime", () => {
+    const { errors, rejected } = report({
+      reason: { kind: "requested-provider-unavailable", providerKey: "vllm" },
+      availableProviderKeys: ["build", "install-vllm"],
+    });
+
+    assert.deepEqual(rejected, []);
+    assert.deepEqual(errors, [
+      "  Requested provider 'vllm' is not available in this environment.",
+      "  NEMOCLAW_PROVIDER=vllm requires an already-running local vLLM server, but none was detected.",
+      "  Re-run with NEMOCLAW_PROVIDER=install-vllm to install and start the managed vLLM runtime.",
+    ]);
+  });
+
+  it("does not recommend managed vLLM when the host has no managed runtime option", () => {
+    const { errors, rejected } = report({
+      reason: { kind: "requested-provider-unavailable", providerKey: "vllm" },
+      availableProviderKeys: ["build"],
+    });
+
+    assert.deepEqual(rejected, []);
+    assert.deepEqual(errors, [
+      "  Requested provider 'vllm' is not available in this environment.",
+      "  NEMOCLAW_PROVIDER=vllm requires an already-running local vLLM server, but none was detected.",
+      "  Start a compatible local vLLM server and retry, or choose another provider.",
     ]);
   });
 });

src/lib/onboard/provider-selection-failure.ts

@@ -5,6 +5,7 @@ import type { ProviderSelectionFailureReason } from "./provider-selection";
 
 export interface ReportProviderSelectionFailureInput {
   reason: ProviderSelectionFailureReason;
+  availableProviderKeys: readonly string[];
   isWindowsHostOllama: boolean;
   rejectWindowsHostOllama(providerKey: string, windowsHostSelected: boolean): boolean;
   writeError(message: string): void;
@@ -44,6 +45,20 @@ export function reportProviderSelectionFailure(input: ReportProviderSelectionFai
       input.writeError(
         `  Requested provider '${input.reason.providerKey}' is not available in this environment.`,
       );
+      if (input.reason.providerKey === "vllm") {
+        input.writeError(
+          "  NEMOCLAW_PROVIDER=vllm requires an already-running local vLLM server, but none was detected.",
+        );
+        input.writeError(
+          input.availableProviderKeys.includes("install-vllm")
+            ? "  Re-run with NEMOCLAW_PROVIDER=install-vllm to install and start the managed vLLM runtime."
+            : "  Start a compatible local vLLM server and retry, or choose another provider.",
+        );
+        break;
+      }
+      input.writeError(
+        "  Re-run without NEMOCLAW_PROVIDER to choose an available provider, or install and start the requested provider before retrying.",
+      );
       break;
   }
 }

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

@@ -618,6 +618,7 @@ export function createSetupNim(
           if (providerSelection.kind === "failure") {
             reportProviderSelectionFailure({
               reason: providerSelection.reason,
+              availableProviderKeys: options.map((option) => option.key),
               isWindowsHostOllama,
               rejectWindowsHostOllama,
               writeError: deps.error,

src/lib/verify-deployment.test.ts

@@ -577,6 +577,32 @@ describe("verifyDeployment", () => {
     expect(sleepCalls).toEqual([10, 10]);
   });
 
+  it("keeps polling through a cold OpenClaw gateway startup before verification (#8901)", async () => {
+    let elapsedMs = 0;
+    const deps = makeDeps({
+      executeSandboxCommand: (_name: string, script: string) => ({
+        status: 0,
+        stdout: script.includes("openclaw --version")
+          ? "2026.5.27"
+          : script.includes("inference.local")
+            ? "200"
+            : elapsedMs >= 60_000
+              ? "200"
+              : "000",
+        stderr: "",
+      }),
+    });
+
+    const result = await verifyDeployment("my-sandbox", chain, deps, {
+      sleep: async (ms: number) => {
+        elapsedMs += ms;
+      },
+    });
+
+    expect(result.healthy).toBe(true);
+    expect(elapsedMs).toBe(60_000);
+  });
+
   it("retries the dashboard probe and recovers when the port forward comes up late (#3563)", async () => {
     let dashboardCalls = 0;
     const deps = makeDeps({

src/lib/verify-deployment.ts

@@ -116,7 +116,7 @@ export interface VerifyDeploymentOptions {
    * returns from createSandbox before the gateway process or the host port
    * forward have finished coming up. Each entry below adds one extra attempt
    * after the initial try, scheduled at the given delay from the previous
-   * attempt. The defaults give roughly a 25 s budget per probe before the
+   * attempt. The defaults give a 90 s budget per probe before the
    * wizard surfaces a ✗ marker.
    * Tests pass `[]` to disable retry.
    */
@@ -131,7 +131,9 @@ export interface VerifyDeploymentOptions {
   diagnoseCustomOpenClawRuntime?: boolean;
 }
 
-const DEFAULT_RETRY_DELAYS_MS: readonly number[] = [1000, 2000, 5000, 7000, 10000];
+const DEFAULT_RETRY_DELAYS_MS: readonly number[] = [
+  1000, 2000, 5000, 7000, 10000, 15000, 20000, 30000,
+];
 // OpenClaw cron stops its provider preflight after 2.5 seconds. Require a
 // response within 2 seconds so onboarding leaves time for client overhead.
 const INFERENCE_ROUTE_REACHABILITY_MAX_SECONDS = 2;

test/managed-image-publication-workflow.test.ts

@@ -656,6 +656,8 @@ describe("complete managed-image publication workflow", () => {
       STAGING_QA_SOURCE_SHA: "250d4abd2d602f864659c06d170699347b5d38bc",
       STAGING_QA_BASE_IMAGE: "nemoclaw-deepagents-code-base:staging-31396519688",
     });
+    expect(qaBuilder.env).not.toHaveProperty("STAGING_PRODUCER_SHA");
+    expect(qaBuilder.env).not.toHaveProperty("STAGING_QA_RECORDED_INDEX_DIGEST");
     expect(JSON.stringify(qaBuilder)).not.toContain(":latest");
     expect(prCheckout.with).toMatchObject({
       ref: "${{ github.event.pull_request.head.sha }}",

test/onboard-selection-windows-provider-rejection.test.ts

@@ -108,6 +108,7 @@ describe("onboard Windows-host Ollama provider rejection", () => {
     const errors: string[] = [];
     reportProviderSelectionFailure({
       reason: failedResolution.reason,
+      availableProviderKeys: options.map((option) => option.key),
       isWindowsHostOllama: false,
       rejectWindowsHostOllama: () => {
         setup();
@@ -143,6 +144,7 @@ describe("onboard Windows-host Ollama provider rejection", () => {
     const errors: string[] = [];
     reportProviderSelectionFailure({
       reason: failedResolution.reason,
+      availableProviderKeys: options.map((option) => option.key),
       isWindowsHostOllama: false,
       rejectWindowsHostOllama: () => {
         install();