chore: add more eval scenarios (#780)

Author: OrKoN

scripts/eval_scenarios/console_test.ts

@@ -0,0 +1,31 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+import assert from 'node:assert';
+
+import type {TestScenario} from '../eval_gemini.ts';
+
+export const scenario: TestScenario = {
+  prompt: 'Navigate to <TEST_URL> and check the console messages.',
+  maxTurns: 2,
+  htmlRoute: {
+    path: '/console_test.html',
+    htmlContent: `
+      <script>
+        console.log('Test log message');
+        console.error('Test error message');
+      </script>
+    `,
+  },
+  expectations: calls => {
+    const navigate = calls.find(
+      c => c.name === 'navigate_page' || c.name === 'new_page',
+    );
+    const listMessages = calls.find(c => c.name === 'list_console_messages');
+
+    assert.ok(navigate, 'Should navigate to the page');
+    assert.ok(listMessages, 'Should list console messages');
+  },
+};

scripts/eval_scenarios/emulation_test.ts

@@ -0,0 +1,18 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+import assert from 'node:assert';
+
+import type {TestScenario} from '../eval_gemini.ts';
+
+export const scenario: TestScenario = {
+  prompt: 'Emulate offline network conditions.',
+  maxTurns: 2,
+  expectations: calls => {
+    const emulate = calls.find(c => c.name === 'emulate');
+    assert.ok(emulate, 'Should call emulate tool');
+    assert.strictEqual(emulate.args.networkConditions, 'Offline');
+  },
+};

scripts/eval_scenarios/input_test.ts

@@ -0,0 +1,38 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+import assert from 'node:assert';
+
+import type {TestScenario} from '../eval_gemini.ts';
+
+export const scenario: TestScenario = {
+  prompt:
+    'Go to <TEST_URL>, fill the input with "hello world" and click the button.',
+  maxTurns: 3,
+  htmlRoute: {
+    path: '/input_test.html',
+    htmlContent: `
+      <input type="text" id="test-input" />
+      <button id="test-button">Submit</button>
+    `,
+  },
+  expectations: calls => {
+    // Expected sequence: navigate -> fill -> click
+    // But model might take snapshot in between or do things in parallel if supported (but standard loop is sequential turns usually)
+    // We just check if the tools were called.
+
+    const navigate = calls.find(
+      c => c.name === 'navigate_page' || c.name === 'new_page',
+    );
+    const fill = calls.find(c => c.name === 'fill');
+    const click = calls.find(c => c.name === 'click');
+
+    assert.ok(navigate, 'Should navigate to the page');
+    assert.ok(fill, 'Should fill the input');
+    assert.ok(click, 'Should click the button');
+
+    assert.strictEqual(fill.args.value, 'hello world');
+  },
+};

scripts/eval_scenarios/network_test.ts

@@ -0,0 +1,31 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+import assert from 'node:assert';
+
+import type {TestScenario} from '../eval_gemini.ts';
+
+export const scenario: TestScenario = {
+  prompt: 'Navigate to <TEST_URL> and list all network requests.',
+  maxTurns: 2,
+  htmlRoute: {
+    path: '/network_test.html',
+    htmlContent: `
+      <h1>Network Test</h1>
+      <script>
+        fetch('/network_test.html'); // Self fetch to ensure at least one request
+      </script>
+    `,
+  },
+  expectations: calls => {
+    const navigate = calls.find(
+      c => c.name === 'navigate_page' || c.name === 'new_page',
+    );
+    const listRequests = calls.find(c => c.name === 'list_network_requests');
+
+    assert.ok(navigate, 'Should navigate to the page');
+    assert.ok(listRequests, 'Should list network requests');
+  },
+};