项目文件夹

文件
Piotr Paulski faac61d2f2 test: fix fill_form eval (#2011)
Improve fill_form eval by checking if fill_form was the only tool called
for the task and provide more descriptive failure messages.

Co-authored-by: Piotr Paulski <piotrpaulski@chromium.org>
2026-05-07 13:59:14 +00:00

89 行
2.7 KiB
TypeScript

/**
* @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 form with size = 2 CPUs and components = [docker, nginx].',
maxTurns: 4, // allow for at least one extra turn to verify there are no extra clicks after fill_form
htmlRoute: {
path: '/input_test.html',
htmlContent: `
<form action="/post" method="POST">
<div>
<label for="size">CPU/Memory size:</label>
<select id="size" name="size" required>
<option value="small">1 vCPU, 2GB RAM</option>
<option value="medium">2 vCPU, 4GB RAM</option>
<option value="large">4 vCPU, 8GB RAM</option>
</select>
</div>
<br>
<div>
<p>Pre-installed components:</p>
<input type="checkbox" id="docker" name="components" value="docker">
<label for="docker">Docker</label><br>
<input type="checkbox" id="nodejs" name="components" value="nodejs">
<label for="nodejs">Node.js</label><br>
<input type="checkbox" id="python" name="components" value="python">
<label for="python">Python</label><br>
<input type="checkbox" id="nginx" name="components" value="nginx">
<label for="nginx">Nginx</label>
</div>
<button type="submit">Spawn Server</button>
</form>
`,
},
expectations: calls => {
assert.ok(calls.length >= 3, 'Not enough calls made');
assert.ok(
calls[0].name === 'navigate_page' || calls[0].name === 'new_page',
);
assert.strictEqual(calls[1].name, 'take_snapshot');
assert.strictEqual(calls[2].name, 'fill_form');
const elements = calls[2].args.elements as Array<{
uid: string;
value: string;
}>;
assert.strictEqual(
elements.length,
3,
'fill_form should be used with all form elements at once',
);
const uids = new Set(elements.map(e => e.uid));
assert.strictEqual(
uids.size,
3,
'fill_form should target three distinct elements',
);
const values = elements.map(e => e.value).sort();
assert.deepStrictEqual(
values,
['2 vCPU, 4GB RAM', 'true', 'true'],
'fill_form should be used with correct values',
);
const submitUid = '1_15';
const extraFormInteractions = calls
.slice(3)
.filter(
c => ['fill', 'click'].includes(c.name) && c.args.uid !== submitUid,
);
assert.deepEqual(
extraFormInteractions.length,
0,
'No extra clicks and fills after fill_form',
);
},
};