test(e2e): add inference adapter fixture (Refs #5745)

.github/workflows/e2e.yaml

@@ -19,6 +19,15 @@ on:
         required: false
         default: ""
         type: string
+      inference_mode:
+        description: "Inference adapter mode for compatible Vitest E2E jobs: mock, internal-nvidia, or public-nvidia."
+        required: false
+        default: "mock"
+        type: choice
+        options:
+          - mock
+          - internal-nvidia
+          - public-nvidia
       allow_jetson_runner_queue:
         description: "Repository administrators only: before setting true for jetson-nvmap-gpu, confirm an online Jetson runner in the authoritative NVIDIA/NemoClaw Settings -> Actions -> Runners inventory; queued jobs do not honor timeout-minutes before assignment."
         required: false
@@ -134,10 +143,15 @@ jobs:
       - id: matrix
         name: Generate E2E target matrix
         env:
+          INFERENCE_MODE: ${{ inputs.inference_mode || 'mock' }}
           JOBS: ${{ inputs.jobs }}
           TARGETS: ${{ inputs.targets }}
         run: |
           set -euo pipefail
+          case "${INFERENCE_MODE}" in
+            mock|internal-nvidia|public-nvidia) ;;
+            *) echo "::error::Invalid inference_mode: ${INFERENCE_MODE}" >&2; exit 1 ;;
+          esac
           if [ -n "${JOBS}" ] && [ -n "${TARGETS}" ]; then
             echo "::error::Use either targets or jobs, not both." >&2
             exit 1
@@ -1674,7 +1688,7 @@ jobs:
       E2E_ARTIFACT_DIR: ${{ github.workspace }}/e2e-artifacts/live/hermes-e2e
       NEMOCLAW_CLI_BIN: ${{ github.workspace }}/bin/nemoclaw.js
       NEMOCLAW_RUN_LIVE_E2E: "1"
-      NEMOCLAW_E2E_USE_HOSTED_INFERENCE: "1"
+      NEMOCLAW_E2E_INFERENCE_MODE: ${{ inputs.inference_mode || 'mock' }}
       NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE: "1"
       NEMOCLAW_AGENT: hermes
       NEMOCLAW_NON_INTERACTIVE: "1"
@@ -1694,7 +1708,7 @@ jobs:
 
       - name: Run Hermes live Vitest test
         env:
-          NVIDIA_INFERENCE_API_KEY: ${{ secrets.NVIDIA_INFERENCE_API_KEY }}
+          NVIDIA_INFERENCE_API_KEY: ${{ github.repository == 'NVIDIA/NemoClaw' && github.ref == 'refs/heads/main' && github.event_name == 'workflow_dispatch' && inputs.checkout_sha == '' && (inputs.inference_mode || 'mock') != 'mock' && secrets.NVIDIA_INFERENCE_API_KEY || '' }}
         run: |
           set -euo pipefail
           npx vitest run --project e2e-live \

test/e2e/fixtures/e2e-test.ts

@@ -13,6 +13,7 @@ import {
   StateClient,
 } from "./clients/index.ts";
 import { DockerPrerequisite, DockerProbe } from "./docker-probe.ts";
+import { createE2EInferenceAdapter, type E2EInferenceAdapter } from "./inference-adapter.ts";
 import {
   EnvironmentPhaseFixture,
   LifecyclePhaseFixture,
@@ -33,6 +34,7 @@ export interface E2ETargetFixtures {
   gateway: GatewayClient;
   sandbox: SandboxClient;
   provider: ProviderClient;
+  inference: E2EInferenceAdapter;
   state: StateClient;
   environment: EnvironmentPhaseFixture;
   onboard: OnboardingPhaseFixture;
@@ -95,6 +97,14 @@ export const test = base.extend<E2ETargetFixtures>({
   provider: async ({ shellProbe }, use) => {
     await use(new ProviderClient(shellProbe));
   },
+  inference: async ({ artifacts, provider, secrets }, use) => {
+    const inference = await createE2EInferenceAdapter({ artifacts, provider, secrets });
+    try {
+      await use(inference);
+    } finally {
+      await inference.close();
+    }
+  },
   state: async ({}, use) => {
     await use(new StateClient());
   },

test/e2e/fixtures/fake-openai-compatible.ts

@@ -26,6 +26,7 @@ export interface FakeOpenAiCompatibleServer {
   readonly baseUrl: string;
   readonly logFile: string;
   readonly requestsFile: string;
+  environmentKeys(): readonly string[];
   requests(): readonly FakeOpenAiCompatibleRequest[];
   close(): Promise<void>;
 }
@@ -119,19 +120,33 @@ function parseRequests(requestsFile: string): FakeOpenAiCompatibleRequest[] {
   }
 }
 
+function parseEnvironmentKeys(environmentFile: string): string[] {
+  try {
+    const parsed = JSON.parse(fs.readFileSync(environmentFile, "utf8")) as unknown;
+    return Array.isArray(parsed)
+      ? parsed.filter((value): value is string => typeof value === "string")
+      : [];
+  } catch {
+    return [];
+  }
+}
+
 export async function startFakeOpenAiCompatibleServer(
   options: FakeOpenAiCompatibleServerOptions = {},
 ): Promise<FakeOpenAiCompatibleServer> {
   const tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), "nemoclaw-fake-openai-"));
+  const environmentFile = path.join(tmpDir, "environment-keys.json");
   const portFile = path.join(tmpDir, "port");
   const logFile = path.join(tmpDir, "server.log");
   const requestsFile = path.join(tmpDir, "requests.jsonl");
   const host = options.host ?? "127.0.0.1";
   const child = spawn(process.execPath, ["--experimental-strip-types", SERVER_SCRIPT], {
     env: {
-      ...process.env,
+      PATH: process.env.PATH ?? "",
+      HOME: process.env.HOME ?? "",
       NEMOCLAW_FAKE_OPENAI_API_KEY: options.apiKey ?? "",
       NEMOCLAW_FAKE_OPENAI_CHAT_CONTENT: options.chatContent ?? "ok",
+      NEMOCLAW_FAKE_OPENAI_ENVIRONMENT_FILE: environmentFile,
       NEMOCLAW_FAKE_OPENAI_FORBIDDEN_MARKERS: JSON.stringify(options.forbiddenMarkers ?? []),
       NEMOCLAW_FAKE_OPENAI_HOST: host,
       NEMOCLAW_FAKE_OPENAI_LOG_FILE: logFile,
@@ -162,6 +177,7 @@ export async function startFakeOpenAiCompatibleServer(
     baseUrl: `http://${formatHttpHost(publicHost)}:${port}/v1`,
     logFile,
     requestsFile,
+    environmentKeys: () => parseEnvironmentKeys(environmentFile),
     requests: () => parseRequests(requestsFile),
     close: async () => {
       child.kill("SIGTERM");