Add browser workflow reliability primitives and guardrails
This commit is contained in:
@@ -8,7 +8,10 @@ const mockUrl = vi.fn().mockReturnValue('https://example.com');
|
||||
const mockClick = vi.fn().mockResolvedValue(undefined);
|
||||
const mockType = vi.fn().mockResolvedValue(undefined);
|
||||
const mock$eval = vi.fn().mockResolvedValue('Page content here');
|
||||
const mock$$eval = vi.fn().mockResolvedValue(['Row 1', 'Row 2']);
|
||||
const mockEvaluate = vi.fn().mockResolvedValue({ result: 42 });
|
||||
const mockWaitForSelector = vi.fn().mockResolvedValue(undefined);
|
||||
const mockWaitForFunction = vi.fn().mockResolvedValue(undefined);
|
||||
const mockScreenshot = vi.fn().mockResolvedValue('base64data');
|
||||
const mock$ = vi.fn().mockResolvedValue({ screenshot: vi.fn().mockResolvedValue('element-base64') });
|
||||
const mockKeyboard = { press: vi.fn().mockResolvedValue(undefined) };
|
||||
@@ -20,7 +23,10 @@ const mockPage = {
|
||||
click: mockClick,
|
||||
type: mockType,
|
||||
$eval: mock$eval,
|
||||
$$eval: mock$$eval,
|
||||
evaluate: mockEvaluate,
|
||||
waitForSelector: mockWaitForSelector,
|
||||
waitForFunction: mockWaitForFunction,
|
||||
screenshot: mockScreenshot,
|
||||
$: mock$,
|
||||
keyboard: mockKeyboard,
|
||||
@@ -44,6 +50,7 @@ describe('Browser tools', () => {
|
||||
|
||||
beforeEach(() => {
|
||||
vi.clearAllMocks();
|
||||
mockUrl.mockReturnValue('https://example.com');
|
||||
tools = createBrowserTools(mockManager);
|
||||
});
|
||||
|
||||
@@ -54,9 +61,14 @@ describe('Browser tools', () => {
|
||||
expect(names).toContain('browser.click');
|
||||
expect(names).toContain('browser.type');
|
||||
expect(names).toContain('browser.content');
|
||||
expect(names).toContain('browser.wait_for');
|
||||
expect(names).toContain('browser.assert');
|
||||
expect(names).toContain('browser.extract');
|
||||
expect(names).toContain('browser.checkpoint.save');
|
||||
expect(names).toContain('browser.checkpoint.resume');
|
||||
expect(names).toContain('browser.eval');
|
||||
expect(names).toContain('browser.evaluate');
|
||||
expect(names).toHaveLength(7);
|
||||
expect(names).toHaveLength(12);
|
||||
});
|
||||
|
||||
it('browser.navigate navigates to URL', async () => {
|
||||
@@ -73,6 +85,17 @@ describe('Browser tools', () => {
|
||||
expect(mockGoto).toHaveBeenCalledWith('https://example.com', { waitUntil: 'networkidle0' });
|
||||
});
|
||||
|
||||
it('browser.navigate retries on transient errors', async () => {
|
||||
mockGoto.mockRejectedValueOnce(new Error('temporary down'));
|
||||
const tool = getTool('browser.navigate');
|
||||
const result = await tool.execute({
|
||||
url: 'https://example.com',
|
||||
retry: { attempts: 2, delay_ms: 0 },
|
||||
});
|
||||
expect(result.success).toBe(true);
|
||||
expect(mockGoto).toHaveBeenCalledTimes(2);
|
||||
});
|
||||
|
||||
it('browser.screenshot takes page screenshot', async () => {
|
||||
const tool = getTool('browser.screenshot');
|
||||
const result = await tool.execute({});
|
||||
@@ -103,6 +126,17 @@ describe('Browser tools', () => {
|
||||
expect(mockClick).toHaveBeenCalledWith('#submit');
|
||||
});
|
||||
|
||||
it('browser.click retries on transient failures', async () => {
|
||||
mockClick.mockRejectedValueOnce(new Error('click miss'));
|
||||
const tool = getTool('browser.click');
|
||||
const result = await tool.execute({
|
||||
selector: '#submit',
|
||||
retry: { attempts: 2, delay_ms: 0 },
|
||||
});
|
||||
expect(result.success).toBe(true);
|
||||
expect(mockClick).toHaveBeenCalledTimes(2);
|
||||
});
|
||||
|
||||
it('browser.type types into element', async () => {
|
||||
const tool = getTool('browser.type');
|
||||
const result = await tool.execute({ selector: '#search', text: 'hello' });
|
||||
@@ -133,6 +167,75 @@ describe('Browser tools', () => {
|
||||
expect(mock$eval).toHaveBeenCalledWith('#main', expect.any(Function));
|
||||
});
|
||||
|
||||
it('browser.wait_for waits on selector and text', async () => {
|
||||
const tool = getTool('browser.wait_for');
|
||||
const result = await tool.execute({
|
||||
selector: '#loaded',
|
||||
text: 'Ready',
|
||||
timeout_ms: 5000,
|
||||
});
|
||||
expect(result.success).toBe(true);
|
||||
expect(mockWaitForSelector).toHaveBeenCalledWith('#loaded', { timeout: 5000, visible: false });
|
||||
expect(mockWaitForFunction).toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it('browser.assert validates selector/text/url conditions', async () => {
|
||||
const tool = getTool('browser.assert');
|
||||
const result = await tool.execute({
|
||||
selector: '#main',
|
||||
exists: true,
|
||||
text: 'Page content here',
|
||||
url_includes: 'example.com',
|
||||
});
|
||||
expect(result.success).toBe(true);
|
||||
});
|
||||
|
||||
it('browser.assert fails when conditions are not met', async () => {
|
||||
mockUrl.mockReturnValue('https://example.com/path');
|
||||
mock$eval.mockResolvedValueOnce('different content');
|
||||
mock$.mockResolvedValueOnce(null);
|
||||
const tool = getTool('browser.assert');
|
||||
const result = await tool.execute({
|
||||
selector: '#missing',
|
||||
exists: true,
|
||||
text: 'not present',
|
||||
url_includes: 'nope',
|
||||
retry: { attempts: 1 },
|
||||
});
|
||||
expect(result.success).toBe(false);
|
||||
expect(result.error).toContain('failed after retries');
|
||||
});
|
||||
|
||||
it('browser.extract returns a single value by selector', async () => {
|
||||
mock$eval.mockResolvedValueOnce('Primary value');
|
||||
const tool = getTool('browser.extract');
|
||||
const result = await tool.execute({ selector: '#value' });
|
||||
expect(result.success).toBe(true);
|
||||
expect(result.output).toContain('"value": "Primary value"');
|
||||
});
|
||||
|
||||
it('browser.extract returns array values when all=true', async () => {
|
||||
mock$$eval.mockResolvedValueOnce(['A', 'B', 'C']);
|
||||
const tool = getTool('browser.extract');
|
||||
const result = await tool.execute({ selector: '.row', all: true });
|
||||
expect(result.success).toBe(true);
|
||||
expect(result.output).toContain('"count": 3');
|
||||
expect(result.output).toContain('"values"');
|
||||
});
|
||||
|
||||
it('browser.checkpoint.save and resume navigates to saved url', async () => {
|
||||
const saveTool = getTool('browser.checkpoint.save');
|
||||
const resumeTool = getTool('browser.checkpoint.resume');
|
||||
|
||||
const saved = await saveTool.execute({ checkpoint_id: 'cp-1' });
|
||||
expect(saved.success).toBe(true);
|
||||
expect(saved.output).toContain('cp-1');
|
||||
|
||||
const resumed = await resumeTool.execute({ checkpoint_id: 'cp-1' });
|
||||
expect(resumed.success).toBe(true);
|
||||
expect(mockGoto).toHaveBeenCalledWith('https://example.com', { waitUntil: 'domcontentloaded' });
|
||||
});
|
||||
|
||||
it('browser.eval evaluates JS', async () => {
|
||||
const tool = getTool('browser.eval');
|
||||
const result = await tool.execute({ expression: '1 + 1' });
|
||||
@@ -155,10 +258,61 @@ describe('Browser tools', () => {
|
||||
expect(result.output).toContain('42');
|
||||
});
|
||||
|
||||
it('enforces allowed domain guardrail for navigation', async () => {
|
||||
const restrictedTools = createBrowserTools(mockManager, {
|
||||
allowedDomains: ['example.com'],
|
||||
});
|
||||
const navigate = restrictedTools.find((tool) => tool.name === 'browser.navigate');
|
||||
if (!navigate) {
|
||||
throw new Error('missing navigate tool');
|
||||
}
|
||||
const result = await navigate.execute({ url: 'https://blocked.test' });
|
||||
expect(result.success).toBe(false);
|
||||
expect(result.error).toContain('allowed_domains');
|
||||
});
|
||||
|
||||
it('requires explicit high-risk confirmation for configured domains', async () => {
|
||||
const guardedTools = createBrowserTools(mockManager, {
|
||||
highRiskDomains: ['bank.example.com'],
|
||||
requireHighRiskConfirmation: true,
|
||||
});
|
||||
const navigate = guardedTools.find((tool) => tool.name === 'browser.navigate');
|
||||
if (!navigate) {
|
||||
throw new Error('missing navigate tool');
|
||||
}
|
||||
|
||||
const denied = await navigate.execute({ url: 'https://bank.example.com' });
|
||||
expect(denied.success).toBe(false);
|
||||
expect(denied.error).toContain('confirm_high_risk=true');
|
||||
|
||||
const allowed = await navigate.execute({
|
||||
url: 'https://bank.example.com',
|
||||
confirm_high_risk: true,
|
||||
});
|
||||
expect(allowed.success).toBe(true);
|
||||
});
|
||||
|
||||
it('enforces workflow step budget', async () => {
|
||||
const budgetedTools = createBrowserTools(mockManager, {
|
||||
maxWorkflowSteps: 1,
|
||||
});
|
||||
const navigate = budgetedTools.find((tool) => tool.name === 'browser.navigate');
|
||||
const click = budgetedTools.find((tool) => tool.name === 'browser.click');
|
||||
if (!navigate || !click) {
|
||||
throw new Error('missing browser tools');
|
||||
}
|
||||
|
||||
const first = await navigate.execute({ url: 'https://example.com' });
|
||||
expect(first.success).toBe(true);
|
||||
const second = await click.execute({ selector: '#submit' });
|
||||
expect(second.success).toBe(false);
|
||||
expect(second.error).toContain('budget exhausted');
|
||||
});
|
||||
|
||||
it('handles navigation errors gracefully', async () => {
|
||||
mockGoto.mockRejectedValueOnce(new Error('Navigation failed'));
|
||||
const tool = getTool('browser.navigate');
|
||||
const result = await tool.execute({ url: 'https://broken.example.com' });
|
||||
const result = await tool.execute({ url: 'https://broken.example.com', retry: { attempts: 1 } });
|
||||
expect(result.success).toBe(false);
|
||||
expect(result.error).toContain('Navigation failed');
|
||||
});
|
||||
@@ -166,7 +320,7 @@ describe('Browser tools', () => {
|
||||
it('handles click errors gracefully', async () => {
|
||||
mockClick.mockRejectedValueOnce(new Error('Element not found'));
|
||||
const tool = getTool('browser.click');
|
||||
const result = await tool.execute({ selector: '#missing' });
|
||||
const result = await tool.execute({ selector: '#missing', retry: { attempts: 1 } });
|
||||
expect(result.success).toBe(false);
|
||||
expect(result.error).toContain('Element not found');
|
||||
});
|
||||
|
||||
+945
-265
File diff suppressed because it is too large
Load Diff
+22
-1
@@ -108,6 +108,11 @@ const PROFILE_TOOLS: Record<ToolProfile, Set<string>> = {
|
||||
'browser.click',
|
||||
'browser.type',
|
||||
'browser.content',
|
||||
'browser.wait_for',
|
||||
'browser.assert',
|
||||
'browser.extract',
|
||||
'browser.checkpoint.save',
|
||||
'browser.checkpoint.resume',
|
||||
'browser.eval',
|
||||
'browser.evaluate',
|
||||
'agent.delegate',
|
||||
@@ -129,7 +134,23 @@ const PROFILE_TOOLS: Record<ToolProfile, Set<string>> = {
|
||||
export const TOOL_GROUPS: Record<string, string[]> = {
|
||||
'group:fs': ['file.read', 'file.write', 'file.edit', 'file.patch', 'file.list'],
|
||||
'group:runtime': ['shell.exec', 'process.start', 'process.output', 'process.status', 'process.kill', 'process.list', 'screen.capture', 'camera.capture'],
|
||||
'group:web': ['web.fetch', 'web.search', 'web.search.news', 'browser.navigate', 'browser.screenshot', 'browser.click', 'browser.type', 'browser.content', 'browser.eval', 'browser.evaluate'],
|
||||
'group:web': [
|
||||
'web.fetch',
|
||||
'web.search',
|
||||
'web.search.news',
|
||||
'browser.navigate',
|
||||
'browser.screenshot',
|
||||
'browser.click',
|
||||
'browser.type',
|
||||
'browser.content',
|
||||
'browser.wait_for',
|
||||
'browser.assert',
|
||||
'browser.extract',
|
||||
'browser.checkpoint.save',
|
||||
'browser.checkpoint.resume',
|
||||
'browser.eval',
|
||||
'browser.evaluate',
|
||||
],
|
||||
'group:memory': ['memory.read', 'memory.write', 'memory.search'],
|
||||
'group:gmail': ['gmail.list', 'gmail.search', 'gmail.read', 'gmail.filter.create'],
|
||||
'group:gcal': ['calendar.today', 'calendar.list', 'calendar.search'],
|
||||
|
||||
Reference in New Issue
Block a user