test(e2e): migrate Hermes inference switch to vitest

.github/workflows/e2e-vitest-scenarios.yaml

@@ -742,6 +742,55 @@ jobs:
           if-no-files-found: ignore
           retention-days: 14
 
+  hermes-inference-switch-vitest:
+    needs: generate-matrix
+    if: ${{ (inputs.jobs == '' && inputs.scenarios == '') || contains(format(',{0},', inputs.jobs), ',hermes-inference-switch-vitest,') || contains(format(',{0},', inputs.scenarios), ',hermes-inference-switch,') }}
+    runs-on: ubuntu-latest
+    timeout-minutes: 55
+    env:
+      FREE_STANDING_VITEST_JOB: "1"
+      FREE_STANDING_SCENARIO_ID: "hermes-inference-switch"
+      E2E_ARTIFACT_DIR: ${{ github.workspace }}/e2e-artifacts/vitest/hermes-inference-switch
+      NEMOCLAW_CLI_BIN: ${{ github.workspace }}/bin/nemoclaw.js
+      NEMOCLAW_RUN_E2E_SCENARIOS: "1"
+      NEMOCLAW_NON_INTERACTIVE: "1"
+      NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE: "1"
+      NEMOCLAW_AGENT: "hermes"
+      NEMOCLAW_SANDBOX_NAME: "e2e-hermes-inference-switch"
+      OPENSHELL_GATEWAY: "nemoclaw"
+    steps:
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        with:
+          persist-credentials: false
+      - name: Set up Node
+        uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e # v6.0.0
+        with:
+          node-version: 22
+          cache: npm
+      - name: Install root dependencies
+        run: npm ci --ignore-scripts
+      - name: Build CLI
+        run: npm run build:cli
+      - name: Install OpenShell CLI
+        run: bash scripts/install-openshell.sh
+      - name: Run Hermes inference switch live Vitest test
+        env:
+          NVIDIA_INFERENCE_API_KEY: ${{ secrets.NVIDIA_INFERENCE_API_KEY }}
+        run: |
+          set -euo pipefail
+          export PATH="$HOME/.local/bin:$HOME/.npm-global/bin:$PATH"
+          export OPENSHELL_BIN="$(command -v openshell || true)"
+          npx vitest run --project e2e-scenarios-live             test/e2e-scenario/live/hermes-inference-switch.test.ts             --silent=false --reporter=default
+      - name: Upload Hermes inference switch artifacts
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: e2e-vitest-scenarios-hermes-inference-switch
+          path: e2e-artifacts/vitest/hermes-inference-switch/
+          include-hidden-files: false
+          if-no-files-found: ignore
+          retention-days: 14
+
   issue-4434-tui-unreachable-inference-vitest:
     needs: generate-matrix
     if: ${{ (inputs.jobs == '' && inputs.scenarios == '') || contains(format(',{0},', inputs.jobs), ',issue-4434-tui-unreachable-inference-vitest,') || contains(format(',{0},', inputs.scenarios), ',issue-4434-tui-unreachable-inference,') }}
@@ -3586,6 +3635,7 @@ jobs:
         openclaw-skill-cli-vitest,
         inference-routing-vitest,
         cloud-inference-vitest,
+        hermes-inference-switch-vitest,
         credential-sanitization-vitest,
         credential-migration-vitest,
         sessions-agents-cli-vitest,

test/e2e-scenario/live/hermes-inference-switch.test.ts

@@ -0,0 +1,260 @@
+// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-License-Identifier: Apache-2.0
+
+/** Live Vitest replacement for test/e2e/test-hermes-inference-switch.sh. */
+
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+import { buildAvailabilityProbeEnv } from "../fixtures/availability-env.ts";
+import { resultText } from "../fixtures/clients/index.ts";
+import { trustedSandboxShellScript, validateSandboxName } from "../fixtures/clients/sandbox.ts";
+import { expect, test } from "../fixtures/e2e-test.ts";
+import { shouldRunLiveE2EScenarios } from "../fixtures/live-project-gate.ts";
+import type { ShellProbeResult } from "../fixtures/shell-probe.ts";
+import { isTransientProviderValidationFailure } from "./network-policy-transient-provider.ts";
+
+const REPO_ROOT = path.resolve(import.meta.dirname, "../../..");
+const CLI = path.join(REPO_ROOT, "bin", "nemoclaw.js");
+const SANDBOX_NAME = process.env.NEMOCLAW_SANDBOX_NAME ?? "e2e-hermes-inference-switch";
+validateSandboxName(SANDBOX_NAME);
+const SWITCH_PROVIDER = process.env.NEMOCLAW_SWITCH_PROVIDER ?? "nvidia-prod";
+const SWITCH_MODEL = process.env.NEMOCLAW_SWITCH_MODEL ?? "z-ai/glm-5.1";
+const SWITCH_API = process.env.NEMOCLAW_SWITCH_INFERENCE_API ?? "openai-completions";
+const INSTALL_ATTEMPTS = process.env.CI === "true" || process.env.GITHUB_ACTIONS === "true" ? 3 : 1;
+const TIMEOUT_MS = 45 * 60_000;
+
+function env(apiKey?: string): NodeJS.ProcessEnv {
+  const out: NodeJS.ProcessEnv = {
+    ...buildAvailabilityProbeEnv(),
+    NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE: "1",
+    NEMOCLAW_AGENT: "hermes",
+    NEMOCLAW_NON_INTERACTIVE: "1",
+    NEMOCLAW_RECREATE_SANDBOX: "1",
+    NEMOCLAW_SANDBOX_NAME: SANDBOX_NAME,
+    OPENSHELL_GATEWAY: process.env.OPENSHELL_GATEWAY ?? "nemoclaw",
+  };
+  if (apiKey) {
+    out.NVIDIA_INFERENCE_API_KEY = apiKey;
+    out.NVIDIA_API_KEY = apiKey;
+  }
+  return out;
+}
+
+async function bestEffort(run: () => Promise<unknown>): Promise<void> {
+  try {
+    await run();
+  } catch {}
+}
+
+function parseHermesModelBlock(text: string): Record<string, string> {
+  const model: Record<string, string> = {};
+  let inModel = false;
+  for (const line of text.split(/\r?\n/u)) {
+    if (/^model:\s*$/u.test(line)) {
+      inModel = true;
+      continue;
+    }
+    if (inModel && /^[A-Za-z0-9_-]+:/u.test(line)) break;
+    if (!inModel) continue;
+    const match = line.match(/^\s+([A-Za-z0-9_-]+):\s*(.*?)\s*$/u);
+    if (!match) continue;
+    const value = match[2].replace(/^['"]|['"]$/gu, "");
+    model[match[1]] = value;
+  }
+  return model;
+}
+
+function chatContent(raw: string): string {
+  const parsed = JSON.parse(raw) as { choices?: Array<{ message?: Record<string, unknown> }> };
+  const message = parsed.choices?.[0]?.message ?? {};
+  for (const key of ["content", "reasoning_content", "reasoning"]) {
+    const value = message[key];
+    if (typeof value === "string" && value.trim()) return value.trim();
+  }
+  return "";
+}
+
+test.skipIf(!shouldRunLiveE2EScenarios())(
+  "Hermes inference set updates route/config and preserves live runtime",
+  { timeout: TIMEOUT_MS },
+  async ({ artifacts, cleanup, host, sandbox, secrets }) => {
+    const apiKey = secrets.required("NVIDIA_INFERENCE_API_KEY");
+    await artifacts.writeJson("scenario.json", {
+      id: "hermes-inference-switch",
+      legacySource: "test/e2e/test-hermes-inference-switch.sh",
+      boundary: "install.sh + Hermes sandbox + inference set + in-sandbox health/chat probes",
+      sandboxName: SANDBOX_NAME,
+      switchProvider: SWITCH_PROVIDER,
+      switchModel: SWITCH_MODEL,
+      switchApi: SWITCH_API,
+    });
+
+    cleanup.add("destroy Hermes inference switch sandbox", async () => {
+      await bestEffort(() =>
+        host.command("node", [CLI, SANDBOX_NAME, "destroy", "--yes"], {
+          artifactName: "cleanup-nemoclaw-destroy",
+          env: env(),
+          timeoutMs: 120_000,
+        }),
+      );
+      await bestEffort(() =>
+        sandbox.openshell(["sandbox", "delete", SANDBOX_NAME], {
+          artifactName: "cleanup-openshell-delete",
+          env: env(),
+          timeoutMs: 60_000,
+        }),
+      );
+    });
+
+    await bestEffort(() =>
+      host.command("node", [CLI, SANDBOX_NAME, "destroy", "--yes"], {
+        artifactName: "pre-cleanup-destroy",
+        env: env(),
+        timeoutMs: 120_000,
+      }),
+    );
+    await bestEffort(() =>
+      sandbox.openshell(["sandbox", "delete", SANDBOX_NAME], {
+        artifactName: "pre-cleanup-delete",
+        env: env(),
+        timeoutMs: 60_000,
+      }),
+    );
+
+    const docker = await host.command("docker", ["info"], {
+      artifactName: "docker-info",
+      env: buildAvailabilityProbeEnv(),
+      timeoutMs: 30_000,
+    });
+    expect(docker.exitCode, resultText(docker)).toBe(0);
+
+    let install: ShellProbeResult | undefined;
+    for (let attempt = 1; attempt <= INSTALL_ATTEMPTS; attempt += 1) {
+      install = await host.command(
+        "bash",
+        ["install.sh", "--non-interactive", "--yes-i-accept-third-party-software"],
+        {
+          artifactName: attempt === 1 ? "install-hermes" : `install-hermes-attempt-${attempt}`,
+          cwd: REPO_ROOT,
+          env: env(apiKey),
+          redactionValues: [apiKey],
+          timeoutMs: 25 * 60_000,
+        },
+      );
+      if (install.exitCode === 0) break;
+      if (isTransientProviderValidationFailure(install) && attempt < INSTALL_ATTEMPTS) {
+        await new Promise((resolve) => setTimeout(resolve, 10_000 * attempt));
+        continue;
+      }
+      break;
+    }
+    expect(install, "install command must run").toBeDefined();
+    expect(install?.exitCode, resultText(install as ShellProbeResult)).toBe(0);
+
+    const pidBefore = await sandbox.execShell(
+      SANDBOX_NAME,
+      trustedSandboxShellScript(
+        "ps -eo pid=,comm=,args= | awk '$0 ~ /hermes/ && $0 ~ /gateway run/ { print $1; exit }'",
+      ),
+      { artifactName: "pid-before", env: env(), timeoutMs: 30_000 },
+    );
+    const envHashBefore = await sandbox.exec(SANDBOX_NAME, ["sha256sum", "/sandbox/.hermes/.env"], {
+      artifactName: "env-hash-before",
+      env: env(),
+      timeoutMs: 30_000,
+    });
+
+    const switched = await host.command(
+      "node",
+      [CLI, "inference", "set", "--provider", SWITCH_PROVIDER, "--model", SWITCH_MODEL],
+      {
+        artifactName: "hermes-inference-set",
+        env: env(apiKey),
+        redactionValues: [apiKey],
+        timeoutMs: 180_000,
+      },
+    );
+    expect(switched.exitCode, resultText(switched)).toBe(0);
+
+    const pidAfter = await sandbox.execShell(
+      SANDBOX_NAME,
+      trustedSandboxShellScript(
+        "ps -eo pid=,comm=,args= | awk '$0 ~ /hermes/ && $0 ~ /gateway run/ { print $1; exit }'",
+      ),
+      { artifactName: "pid-after", env: env(), timeoutMs: 30_000 },
+    );
+    if (pidBefore.stdout.trim() && pidAfter.stdout.trim())
+      expect(pidAfter.stdout.trim()).toBe(pidBefore.stdout.trim());
+
+    const health = await sandbox.exec(
+      SANDBOX_NAME,
+      ["curl", "-sf", "--max-time", "10", "http://localhost:8642/health"],
+      { artifactName: "hermes-health-after-switch", env: env(), timeoutMs: 30_000 },
+    );
+    expect(health.exitCode, resultText(health)).toBe(0);
+    expect(resultText(health)).toMatch(/ok/i);
+
+    const route = await sandbox.openshell(["inference", "get", "-g", "nemoclaw"], {
+      artifactName: "openshell-inference-route",
+      env: env(),
+      timeoutMs: 30_000,
+    });
+    expect(route.exitCode, resultText(route)).toBe(0);
+    expect(resultText(route)).toContain(SWITCH_PROVIDER);
+    expect(resultText(route)).toContain(SWITCH_MODEL);
+
+    const config = await sandbox.exec(SANDBOX_NAME, ["cat", "/sandbox/.hermes/config.yaml"], {
+      artifactName: "hermes-config-yaml",
+      env: env(),
+      redactionValues: [apiKey],
+      timeoutMs: 30_000,
+    });
+    expect(config.exitCode, resultText(config)).toBe(0);
+    const model = parseHermesModelBlock(config.stdout);
+    expect(model.default).toBe(SWITCH_MODEL);
+    expect(model.provider).toBe("custom");
+    expect(model.base_url).toBe(
+      SWITCH_API === "anthropic-messages"
+        ? "https://inference.local"
+        : "https://inference.local/v1",
+    );
+    expect(config.stdout).not.toMatch(/^models:\s*$/mu);
+
+    const envHashAfter = await sandbox.exec(SANDBOX_NAME, ["sha256sum", "/sandbox/.hermes/.env"], {
+      artifactName: "env-hash-after",
+      env: env(),
+      timeoutMs: 30_000,
+    });
+    if (envHashBefore.stdout.trim())
+      expect(envHashAfter.stdout.split(/\s+/u)[0]).toBe(envHashBefore.stdout.split(/\s+/u)[0]);
+
+    const registry = JSON.parse(
+      fs.readFileSync(path.join(os.homedir(), ".nemoclaw", "sandboxes.json"), "utf8"),
+    );
+    expect(registry.sandboxes?.[SANDBOX_NAME]?.agent).toBe("hermes");
+    expect(registry.sandboxes?.[SANDBOX_NAME]?.provider).toBe(SWITCH_PROVIDER);
+    expect(registry.sandboxes?.[SANDBOX_NAME]?.model).toBe(SWITCH_MODEL);
+
+    const payload = JSON.stringify({
+      model: SWITCH_MODEL,
+      messages: [{ role: "user", content: "Reply with exactly one word: PONG" }],
+      max_tokens: 100,
+    });
+    const chat = await sandbox.execShell(
+      SANDBOX_NAME,
+      trustedSandboxShellScript(
+        `set -a; [ ! -f /sandbox/.hermes/.env ] || . /sandbox/.hermes/.env; set +a; curl -sS --max-time 120 http://localhost:8642/v1/chat/completions -H 'Content-Type: application/json' -H "Authorization: Bearer \${API_SERVER_KEY:-}" -d '${payload.replace(/'/gu, `'\\''`)}'`,
+      ),
+      {
+        artifactName: "hermes-api-chat-after-switch",
+        env: env(),
+        redactionValues: [apiKey],
+        timeoutMs: 150_000,
+      },
+    );
+    expect(chat.exitCode, resultText(chat)).toBe(0);
+    expect(chatContent(chat.stdout)).toMatch(/PONG/i);
+  },
+);