@loadmill/droid-cua 2.21.0 → 2.23.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/build/index.js +31 -1
- package/build/src/browser/playwright/actions.js +65 -1
- package/build/src/cli/headless-runner.js +174 -12
- package/build/src/commands/help.js +1 -1
- package/build/src/core/cua-models.js +11 -4
- package/build/src/core/droid-cua-run-usage.js +106 -0
- package/build/src/core/droid-cua-segment-events.js +32 -0
- package/build/src/core/execution-engine.js +334 -5
- package/build/src/core/prompts.js +1 -1
- package/build/src/core/report/render-html-report.js +8 -3
- package/build/src/core/report/run-log-model.js +42 -0
- package/build/src/core/runtime-config.js +24 -39
- package/build/src/core/session.js +6 -0
- package/build/src/core/tool-handlers/loadmill.js +3 -0
- package/build/src/core/tool-handlers/test-design-complete.js +1 -0
- package/build/src/core/tool-handlers/type-sensitive-text.js +1 -0
- package/build/src/device/android/actions.js +6 -2
- package/build/src/device/android/connection.js +58 -23
- package/build/src/device/assertions.js +15 -8
- package/build/src/device/ios/connection.js +36 -30
- package/build/src/device/openai.js +68 -2
- package/build/src/device/scroll-gesture.js +22 -9
- package/build/src/integrations/loadmill/client.js +62 -0
- package/build/src/integrations/loadmill/run-reporter.js +113 -0
- package/build/src/integrations/loadmill/session-user.js +39 -0
- package/build/src/integrations/segment/droid-cua-segment.js +64 -0
- package/build/src/modes/design-mode-ink.js +10 -1
- package/build/src/modes/design-mode.js +7 -1
- package/build/src/modes/execution-mode.js +84 -8
- package/build/src/prompts/design.js +21 -1
- package/build/src/utils/console-output.js +5 -0
- package/build/src/utils/cua-debug-tracer.js +50 -7
- package/package.json +2 -1
- package/build/src/core/cua-backends.js +0 -11
package/build/index.js
CHANGED
|
@@ -15,7 +15,7 @@ import { selectDevice } from "./src/cli/device-selector.js";
|
|
|
15
15
|
import { resolveExecutionAppContext } from "./src/core/app-context.js";
|
|
16
16
|
import { resolveExecutionTestData } from "./src/core/test-data.js";
|
|
17
17
|
import { getDefaultSecretsPath, resolveExecutionSecrets } from "./src/core/secrets.js";
|
|
18
|
-
import { createRunContext, resolveRuntimeConfigFromEnv, resolveRuntimeCredentialsFromEnv, } from "./src/core/runtime-config.js";
|
|
18
|
+
import { assertRuntimeCredentialsForProvider, createRunContext, resolveRuntimeConfigFromEnv, resolveRuntimeCredentialsFromEnv, } from "./src/core/runtime-config.js";
|
|
19
19
|
import { resolveHeadlessExecutionConfig } from "./src/cli/headless-execution-config.js";
|
|
20
20
|
import { resolveHeadlessTarget } from "./src/cli/headless-target.js";
|
|
21
21
|
import { createHeadlessDebugArtifacts } from "./src/cli/headless-debug.js";
|
|
@@ -71,6 +71,33 @@ function createCliRuntimeContext(configOverrides = {}) {
|
|
|
71
71
|
loadmillClient,
|
|
72
72
|
};
|
|
73
73
|
}
|
|
74
|
+
async function assertCliRuntimeCredentials({ llmProviderMode }) {
|
|
75
|
+
const credentials = resolveRuntimeCredentialsFromEnv(process.env);
|
|
76
|
+
assertRuntimeCredentialsForProvider({
|
|
77
|
+
llmProviderMode,
|
|
78
|
+
credentials,
|
|
79
|
+
});
|
|
80
|
+
if (llmProviderMode !== "loadmill") {
|
|
81
|
+
return;
|
|
82
|
+
}
|
|
83
|
+
const loadmillClient = createLoadmillApiClient({
|
|
84
|
+
token: credentials.loadmillApiToken,
|
|
85
|
+
baseUrl: credentials.loadmillBaseUrl,
|
|
86
|
+
});
|
|
87
|
+
let tokenValidation = null;
|
|
88
|
+
try {
|
|
89
|
+
tokenValidation = await loadmillClient.validateApiToken();
|
|
90
|
+
}
|
|
91
|
+
catch {
|
|
92
|
+
throw new Error("Could not validate LOADMILL_API_TOKEN.");
|
|
93
|
+
}
|
|
94
|
+
if (tokenValidation?.reason === "invalid") {
|
|
95
|
+
throw new Error("LOADMILL_API_TOKEN is invalid.");
|
|
96
|
+
}
|
|
97
|
+
if (!tokenValidation?.valid) {
|
|
98
|
+
throw new Error("Could not validate LOADMILL_API_TOKEN.");
|
|
99
|
+
}
|
|
100
|
+
}
|
|
74
101
|
function createLambdaTestCliBuildName() {
|
|
75
102
|
return `loadmill-droid-cua-cli-${new Date().toISOString().replace(/[:.]/g, "-")}`;
|
|
76
103
|
}
|
|
@@ -184,6 +211,9 @@ async function main() {
|
|
|
184
211
|
let preloadedSecrets = null;
|
|
185
212
|
if (isHeadlessInstructionsRun) {
|
|
186
213
|
preloadedHeadlessConfig = await resolveHeadlessExecutionConfig(args);
|
|
214
|
+
await assertCliRuntimeCredentials({
|
|
215
|
+
llmProviderMode: preloadedHeadlessConfig.llmProviderMode,
|
|
216
|
+
});
|
|
187
217
|
preloadedHeadlessTarget = await resolveHeadlessInstructionsTarget(instructionsFile);
|
|
188
218
|
preloadedSecrets = await resolveExecutionSecrets({
|
|
189
219
|
secretsPath: getDefaultSecretsPath(process.cwd()),
|
|
@@ -3,7 +3,8 @@ import { emitDesktopDebug, truncateForDebug } from "../../utils/desktop-debug.js
|
|
|
3
3
|
import { getConfiguredStepDelayMs } from "../../utils/step-delay.js";
|
|
4
4
|
import { buildActionEventMeta } from "../../device/action-event-meta.js";
|
|
5
5
|
import { actionForErrorLog, buildActionErrorOutput, buildDebugTextField, buildSensitiveDebugFields, buildTypeActionPayload, buildTypeActionText, isSensitiveAction, } from "../../device/sensitive-action.js";
|
|
6
|
-
import {
|
|
6
|
+
import { normalizeAutomationUrl } from "./url.js";
|
|
7
|
+
import { getActiveBrowserSession, getActiveBrowserPage } from "./session-state.js";
|
|
7
8
|
const KEY_ALIASES = {
|
|
8
9
|
CTRL: "Control",
|
|
9
10
|
CONTROL: "Control",
|
|
@@ -58,8 +59,47 @@ function buildMeta(context, action, payload = {}) {
|
|
|
58
59
|
payload,
|
|
59
60
|
});
|
|
60
61
|
}
|
|
62
|
+
function handleNavigationTimeout(error, description) {
|
|
63
|
+
if (!(error instanceof Error) || error.name !== "TimeoutError") {
|
|
64
|
+
throw error;
|
|
65
|
+
}
|
|
66
|
+
console.warn(`Page load did not finish within the timeout for ${description}. Continuing with the current page state.`);
|
|
67
|
+
}
|
|
68
|
+
async function gotoWithTimeoutTolerance(page, url) {
|
|
69
|
+
try {
|
|
70
|
+
await page.goto(url, { waitUntil: "load" });
|
|
71
|
+
}
|
|
72
|
+
catch (error) {
|
|
73
|
+
handleNavigationTimeout(error, url);
|
|
74
|
+
}
|
|
75
|
+
}
|
|
76
|
+
async function goBackWithTimeoutTolerance(page) {
|
|
77
|
+
try {
|
|
78
|
+
await page.goBack({ waitUntil: "load" });
|
|
79
|
+
}
|
|
80
|
+
catch (error) {
|
|
81
|
+
handleNavigationTimeout(error, "browser history back");
|
|
82
|
+
}
|
|
83
|
+
}
|
|
84
|
+
async function goForwardWithTimeoutTolerance(page) {
|
|
85
|
+
try {
|
|
86
|
+
await page.goForward({ waitUntil: "load" });
|
|
87
|
+
}
|
|
88
|
+
catch (error) {
|
|
89
|
+
handleNavigationTimeout(error, "browser history forward");
|
|
90
|
+
}
|
|
91
|
+
}
|
|
92
|
+
function updateSessionUrlFromPage(session, page) {
|
|
93
|
+
const currentUrl = typeof page.url === "function" ? page.url() : null;
|
|
94
|
+
if (!currentUrl) {
|
|
95
|
+
return;
|
|
96
|
+
}
|
|
97
|
+
session.pageUrls?.set?.(page, currentUrl);
|
|
98
|
+
session.url = currentUrl;
|
|
99
|
+
}
|
|
61
100
|
export async function handleModelAction(browserSessionId, action, scale = 1, context = null) {
|
|
62
101
|
const addOutput = context?.addOutput || ((item) => console.log(item.text || item));
|
|
102
|
+
const session = getActiveBrowserSession(browserSessionId);
|
|
63
103
|
const page = getActiveBrowserPage(browserSessionId);
|
|
64
104
|
try {
|
|
65
105
|
emitDesktopDebug("browser.action.execute", "device", {
|
|
@@ -166,6 +206,30 @@ export async function handleModelAction(browserSessionId, action, scale = 1, con
|
|
|
166
206
|
await page.waitForTimeout(timeoutMs);
|
|
167
207
|
break;
|
|
168
208
|
}
|
|
209
|
+
case "goto": {
|
|
210
|
+
const normalizedUrl = normalizeAutomationUrl(action.url);
|
|
211
|
+
addOutput({
|
|
212
|
+
type: "action",
|
|
213
|
+
text: `Navigating to ${normalizedUrl}`,
|
|
214
|
+
...buildMeta(context, action, { url: normalizedUrl }),
|
|
215
|
+
});
|
|
216
|
+
await gotoWithTimeoutTolerance(page, normalizedUrl);
|
|
217
|
+
session.pageUrls?.set?.(page, normalizedUrl);
|
|
218
|
+
session.url = normalizedUrl;
|
|
219
|
+
break;
|
|
220
|
+
}
|
|
221
|
+
case "back": {
|
|
222
|
+
addOutput({ type: "action", text: "Going back", ...buildMeta(context, action) });
|
|
223
|
+
await goBackWithTimeoutTolerance(page);
|
|
224
|
+
updateSessionUrlFromPage(session, page);
|
|
225
|
+
break;
|
|
226
|
+
}
|
|
227
|
+
case "forward": {
|
|
228
|
+
addOutput({ type: "action", text: "Going forward", ...buildMeta(context, action) });
|
|
229
|
+
await goForwardWithTimeoutTolerance(page);
|
|
230
|
+
updateSessionUrlFromPage(session, page);
|
|
231
|
+
break;
|
|
232
|
+
}
|
|
169
233
|
default:
|
|
170
234
|
addOutput({ type: "info", text: `Unknown action: ${JSON.stringify(action)}` });
|
|
171
235
|
}
|
|
@@ -4,7 +4,10 @@ import { getCurrentPlatform as getBrowserCurrentPlatform, getScreenshotAsBase64
|
|
|
4
4
|
import { handleModelAction as handleBrowserModelAction } from "../browser/actions.js";
|
|
5
5
|
import { ExecutionEngine } from "../core/execution-engine.js";
|
|
6
6
|
import { buildExecutionModePrompt } from "../core/prompts.js";
|
|
7
|
+
import { addOpenAiResponseToDroidCuaRunUsage, createDroidCuaRunUsageTotals, estimateDroidCuaRunCostUsd, setDroidCuaRunUsageModel, } from "../core/droid-cua-run-usage.js";
|
|
7
8
|
import { ExecutionMode } from "../modes/execution-mode.js";
|
|
9
|
+
import { reportDroidRunCompletion } from "../integrations/loadmill/run-reporter.js";
|
|
10
|
+
import { reportDroidCuaCiTestRunCompleted } from "../integrations/segment/droid-cua-segment.js";
|
|
8
11
|
import { printCliOutput } from "../utils/console-output.js";
|
|
9
12
|
import { emitDesktopDebug } from "../utils/desktop-debug.js";
|
|
10
13
|
import { loadHeadlessInstructions } from "./headless-instructions.js";
|
|
@@ -27,6 +30,18 @@ function getRecordScreenshotDirForTest(baseScreenshotDir, testName, isBatchRun)
|
|
|
27
30
|
}
|
|
28
31
|
return path.join(baseScreenshotDir, sanitizePathSegment(testName));
|
|
29
32
|
}
|
|
33
|
+
function toDroidCuaSegmentRunStatus(success, error = null) {
|
|
34
|
+
if (success) {
|
|
35
|
+
return "pass";
|
|
36
|
+
}
|
|
37
|
+
if (error === "Stopped by user") {
|
|
38
|
+
return "skipped";
|
|
39
|
+
}
|
|
40
|
+
if (typeof error === "string" && error.startsWith("Assertion failed:")) {
|
|
41
|
+
return "failure";
|
|
42
|
+
}
|
|
43
|
+
return "error";
|
|
44
|
+
}
|
|
30
45
|
function emitCliAppContextDebug(runId, appContext, appContextText) {
|
|
31
46
|
if (appContext.status === "loaded" || appContext.status === "empty") {
|
|
32
47
|
emitDesktopDebug("app_context.loaded", "execution", { runId }, {
|
|
@@ -56,7 +71,17 @@ function emitCliAppContextDebug(runId, appContext, appContextText) {
|
|
|
56
71
|
});
|
|
57
72
|
}
|
|
58
73
|
}
|
|
59
|
-
function buildHeadlessEngineOptions({ runtimeContext, headlessConfig, executionPrompt, screenshotRecorder, recordScreenshots, screenshotDir, isBrowserTarget, isCloudTarget, secrets = null }) {
|
|
74
|
+
function buildHeadlessEngineOptions({ runtimeContext, headlessConfig, executionPrompt, screenshotRecorder, recordScreenshots, screenshotDir, isBrowserTarget, isCloudTarget, onCuaResponse = null, secrets = null }) {
|
|
75
|
+
const sendCUARequest = async (requestArgs) => {
|
|
76
|
+
const response = await runtimeContext.openAi.sendCUARequest({
|
|
77
|
+
...requestArgs,
|
|
78
|
+
aimAssistMode: headlessConfig.aimAssistMode,
|
|
79
|
+
});
|
|
80
|
+
if (typeof onCuaResponse === "function") {
|
|
81
|
+
onCuaResponse(response);
|
|
82
|
+
}
|
|
83
|
+
return response;
|
|
84
|
+
};
|
|
60
85
|
const executionEngineOptions = {
|
|
61
86
|
toolCapabilities: {
|
|
62
87
|
assert: true,
|
|
@@ -72,10 +97,7 @@ function buildHeadlessEngineOptions({ runtimeContext, headlessConfig, executionP
|
|
|
72
97
|
recordScreenshots: true,
|
|
73
98
|
screenshotRecorder,
|
|
74
99
|
stepDelayMs: runtimeContext.config.stepDelayMs,
|
|
75
|
-
sendCUARequest
|
|
76
|
-
...requestArgs,
|
|
77
|
-
aimAssistMode: headlessConfig.aimAssistMode,
|
|
78
|
-
}),
|
|
100
|
+
sendCUARequest,
|
|
79
101
|
reviseTestScript: runtimeContext.openAi.reviseTestScript,
|
|
80
102
|
...executionEngineOptions,
|
|
81
103
|
}
|
|
@@ -83,10 +105,7 @@ function buildHeadlessEngineOptions({ runtimeContext, headlessConfig, executionP
|
|
|
83
105
|
recordScreenshots,
|
|
84
106
|
screenshotDir,
|
|
85
107
|
stepDelayMs: runtimeContext.config.stepDelayMs,
|
|
86
|
-
sendCUARequest
|
|
87
|
-
...requestArgs,
|
|
88
|
-
aimAssistMode: headlessConfig.aimAssistMode,
|
|
89
|
-
}),
|
|
108
|
+
sendCUARequest,
|
|
90
109
|
reviseTestScript: runtimeContext.openAi.reviseTestScript,
|
|
91
110
|
...executionEngineOptions,
|
|
92
111
|
};
|
|
@@ -102,14 +121,38 @@ function buildHeadlessEngineOptions({ runtimeContext, headlessConfig, executionP
|
|
|
102
121
|
}
|
|
103
122
|
return engineOptions;
|
|
104
123
|
}
|
|
105
|
-
async function runHeadlessTest({ runId, testPath, testName, instructions, session, deviceInfo, headlessConfig, appContext, testDataContext = "", runtimeContext, headlessDebug, recordScreenshots, screenshotDir, isBrowserTarget, isCloudTarget, secrets = null, report = null, isBatchRun = false }) {
|
|
124
|
+
async function runHeadlessTest({ runId, testPath, testName, instructions, session, deviceInfo, headlessConfig, appContext, testDataContext = "", runtimeContext, headlessDebug, recordScreenshots, screenshotDir, isBrowserTarget, isCloudTarget, secrets = null, report = null, isBatchRun = false, reportToLoadmill = true }) {
|
|
106
125
|
let executionMode = null;
|
|
126
|
+
let loadmillSessionUser = null;
|
|
127
|
+
let runUsageTotals = setDroidCuaRunUsageModel(createDroidCuaRunUsageTotals(), runtimeContext.config.cuaModel);
|
|
128
|
+
const buildSegmentUsageProperties = () => ({
|
|
129
|
+
model: runUsageTotals.model,
|
|
130
|
+
llmProviderMode: runtimeContext.config.llmProviderMode,
|
|
131
|
+
estimatedCostUsd: estimateDroidCuaRunCostUsd(runUsageTotals),
|
|
132
|
+
tokensUsed: runUsageTotals.totalTokens,
|
|
133
|
+
});
|
|
134
|
+
const reportSegmentRunCompleted = async (input) => {
|
|
135
|
+
await reportDroidCuaCiTestRunCompleted({
|
|
136
|
+
...input,
|
|
137
|
+
runId,
|
|
138
|
+
platform: deviceInfo.platform,
|
|
139
|
+
userId: loadmillSessionUser?.id ?? null,
|
|
140
|
+
teamId: loadmillSessionUser?.teamId ?? null,
|
|
141
|
+
...buildSegmentUsageProperties(),
|
|
142
|
+
});
|
|
143
|
+
};
|
|
107
144
|
const reportEnabled = Boolean(report?.enabled);
|
|
108
145
|
const collectedEvents = [];
|
|
109
146
|
const addOutput = reportEnabled
|
|
110
147
|
? createCapturingAddOutput(printCliOutput, collectedEvents)
|
|
111
148
|
: printCliOutput;
|
|
112
149
|
try {
|
|
150
|
+
try {
|
|
151
|
+
loadmillSessionUser = await runtimeContext.loadmillClient?.getCurrentUser?.();
|
|
152
|
+
}
|
|
153
|
+
catch {
|
|
154
|
+
loadmillSessionUser = null;
|
|
155
|
+
}
|
|
113
156
|
await headlessDebug.startExecutionSession(runId, {
|
|
114
157
|
testName,
|
|
115
158
|
...(testDataContext ? { testDataUsed: true } : {}),
|
|
@@ -147,6 +190,9 @@ async function runHeadlessTest({ runId, testPath, testName, instructions, sessio
|
|
|
147
190
|
screenshotDir,
|
|
148
191
|
isBrowserTarget,
|
|
149
192
|
isCloudTarget,
|
|
193
|
+
onCuaResponse: (response) => {
|
|
194
|
+
runUsageTotals = addOpenAiResponseToDroidCuaRunUsage(runUsageTotals, response);
|
|
195
|
+
},
|
|
150
196
|
secrets
|
|
151
197
|
});
|
|
152
198
|
const engine = new ExecutionEngine(session, engineOptions);
|
|
@@ -174,6 +220,37 @@ async function runHeadlessTest({ runId, testPath, testName, instructions, sessio
|
|
|
174
220
|
assertionsFailed: stats.assertionsFailed ?? 0,
|
|
175
221
|
retries: stats.retryCount ?? 0
|
|
176
222
|
});
|
|
223
|
+
await reportSegmentRunCompleted({
|
|
224
|
+
status: toDroidCuaSegmentRunStatus(Boolean(result.success), result.error ?? null),
|
|
225
|
+
durationMs,
|
|
226
|
+
numOfSteps: instructions.length,
|
|
227
|
+
});
|
|
228
|
+
if (reportToLoadmill) {
|
|
229
|
+
await reportDroidRunCompletion({
|
|
230
|
+
loadmillClient: runtimeContext.loadmillClient,
|
|
231
|
+
source: "ci",
|
|
232
|
+
runId,
|
|
233
|
+
projectName: path.basename(path.dirname(testPath)) || null,
|
|
234
|
+
testName,
|
|
235
|
+
runType: "test",
|
|
236
|
+
testCases: 1,
|
|
237
|
+
success: Boolean(result.success),
|
|
238
|
+
error: result.error ?? null,
|
|
239
|
+
completedAt: new Date().toISOString(),
|
|
240
|
+
durationMs,
|
|
241
|
+
platform: deviceInfo.platform,
|
|
242
|
+
deviceSource: deviceInfo.source,
|
|
243
|
+
deviceName: deviceInfo.device_name,
|
|
244
|
+
osVersion: deviceInfo.os_version,
|
|
245
|
+
instructionsTotal: instructions.length,
|
|
246
|
+
instructionsCompleted: stats.instructionsCompleted ?? 0,
|
|
247
|
+
actionsTotal: stats.actionCount ?? 0,
|
|
248
|
+
assertionsPassed: stats.assertionsPassed ?? 0,
|
|
249
|
+
assertionsFailed: stats.assertionsFailed ?? 0,
|
|
250
|
+
retries: stats.retryCount ?? 0,
|
|
251
|
+
onWarning: (message) => console.warn(`Warning: ${message}`),
|
|
252
|
+
});
|
|
253
|
+
}
|
|
177
254
|
if (!result.success) {
|
|
178
255
|
console.error(`\nTest failed: ${result.error}`);
|
|
179
256
|
}
|
|
@@ -199,7 +276,14 @@ async function runHeadlessTest({ runId, testPath, testName, instructions, sessio
|
|
|
199
276
|
}
|
|
200
277
|
return {
|
|
201
278
|
success: Boolean(result.success),
|
|
202
|
-
error: result.error ?? null
|
|
279
|
+
error: result.error ?? null,
|
|
280
|
+
durationMs,
|
|
281
|
+
instructionsTotal: instructions.length,
|
|
282
|
+
instructionsCompleted: stats.instructionsCompleted ?? 0,
|
|
283
|
+
actionsTotal: stats.actionCount ?? 0,
|
|
284
|
+
assertionsPassed: stats.assertionsPassed ?? 0,
|
|
285
|
+
assertionsFailed: stats.assertionsFailed ?? 0,
|
|
286
|
+
retries: stats.retryCount ?? 0
|
|
203
287
|
};
|
|
204
288
|
}
|
|
205
289
|
catch (error) {
|
|
@@ -221,6 +305,37 @@ async function runHeadlessTest({ runId, testPath, testName, instructions, sessio
|
|
|
221
305
|
}
|
|
222
306
|
: { reason: "start_failed" })
|
|
223
307
|
});
|
|
308
|
+
await reportSegmentRunCompleted({
|
|
309
|
+
status: "error",
|
|
310
|
+
durationMs,
|
|
311
|
+
numOfSteps: instructions.length,
|
|
312
|
+
});
|
|
313
|
+
if (reportToLoadmill) {
|
|
314
|
+
await reportDroidRunCompletion({
|
|
315
|
+
loadmillClient: runtimeContext.loadmillClient,
|
|
316
|
+
source: "ci",
|
|
317
|
+
runId,
|
|
318
|
+
projectName: path.basename(path.dirname(testPath)) || null,
|
|
319
|
+
testName,
|
|
320
|
+
runType: "test",
|
|
321
|
+
testCases: 1,
|
|
322
|
+
success: false,
|
|
323
|
+
error: message,
|
|
324
|
+
completedAt: new Date().toISOString(),
|
|
325
|
+
durationMs,
|
|
326
|
+
platform: deviceInfo.platform,
|
|
327
|
+
deviceSource: deviceInfo.source,
|
|
328
|
+
deviceName: deviceInfo.device_name,
|
|
329
|
+
osVersion: deviceInfo.os_version,
|
|
330
|
+
instructionsTotal: instructions.length,
|
|
331
|
+
instructionsCompleted: stats.instructionsCompleted ?? 0,
|
|
332
|
+
actionsTotal: stats.actionCount ?? 0,
|
|
333
|
+
assertionsPassed: stats.assertionsPassed ?? 0,
|
|
334
|
+
assertionsFailed: stats.assertionsFailed ?? 0,
|
|
335
|
+
retries: stats.retryCount ?? 0,
|
|
336
|
+
onWarning: (warningMessage) => console.warn(`Warning: ${warningMessage}`),
|
|
337
|
+
});
|
|
338
|
+
}
|
|
224
339
|
throw error;
|
|
225
340
|
}
|
|
226
341
|
}
|
|
@@ -233,6 +348,16 @@ export async function runHeadlessInstructionRuns({ runs, instructionsTarget, ses
|
|
|
233
348
|
console.log(`\nRunning ${runs.length} tests from: ${instructionsTarget.path}\n`);
|
|
234
349
|
}
|
|
235
350
|
let failures = 0;
|
|
351
|
+
const batchStartedAt = Date.now();
|
|
352
|
+
const aggregateStats = {
|
|
353
|
+
durationMs: 0,
|
|
354
|
+
instructionsTotal: 0,
|
|
355
|
+
instructionsCompleted: 0,
|
|
356
|
+
actionsTotal: 0,
|
|
357
|
+
assertionsPassed: 0,
|
|
358
|
+
assertionsFailed: 0,
|
|
359
|
+
retries: 0
|
|
360
|
+
};
|
|
236
361
|
for (const [index, run] of runs.entries()) {
|
|
237
362
|
const runId = createHeadlessRunId(index);
|
|
238
363
|
const testScreenshotDir = getRecordScreenshotDirForTest(screenshotDir, run.testName, isBatchRun);
|
|
@@ -262,8 +387,16 @@ export async function runHeadlessInstructionRuns({ runs, instructionsTarget, ses
|
|
|
262
387
|
isBrowserTarget,
|
|
263
388
|
isCloudTarget,
|
|
264
389
|
report,
|
|
265
|
-
isBatchRun
|
|
390
|
+
isBatchRun,
|
|
391
|
+
reportToLoadmill: !isBatchRun
|
|
266
392
|
});
|
|
393
|
+
aggregateStats.durationMs += result.durationMs ?? 0;
|
|
394
|
+
aggregateStats.instructionsTotal += result.instructionsTotal ?? 0;
|
|
395
|
+
aggregateStats.instructionsCompleted += result.instructionsCompleted ?? 0;
|
|
396
|
+
aggregateStats.actionsTotal += result.actionsTotal ?? 0;
|
|
397
|
+
aggregateStats.assertionsPassed += result.assertionsPassed ?? 0;
|
|
398
|
+
aggregateStats.assertionsFailed += result.assertionsFailed ?? 0;
|
|
399
|
+
aggregateStats.retries += result.retries ?? 0;
|
|
267
400
|
if (!result.success) {
|
|
268
401
|
failures += 1;
|
|
269
402
|
}
|
|
@@ -285,6 +418,35 @@ export async function runHeadlessInstructionRuns({ runs, instructionsTarget, ses
|
|
|
285
418
|
const passed = runs.length - failures;
|
|
286
419
|
console.log(`\nFolder run complete: ${passed}/${runs.length} passed`);
|
|
287
420
|
}
|
|
421
|
+
if (isBatchRun) {
|
|
422
|
+
const isScenario = instructionsTarget?.type === "scenario";
|
|
423
|
+
const testName = isScenario ? "Scenario run" : path.basename(instructionsTarget.path);
|
|
424
|
+
await reportDroidRunCompletion({
|
|
425
|
+
loadmillClient: runtimeContext.loadmillClient,
|
|
426
|
+
source: "ci",
|
|
427
|
+
runId: createHeadlessRunId(),
|
|
428
|
+
projectName: isScenario ? path.basename(process.cwd()) : path.basename(instructionsTarget.path),
|
|
429
|
+
testName,
|
|
430
|
+
runType: "project",
|
|
431
|
+
testCases: runs.length,
|
|
432
|
+
outcome: failures > 0 ? "failure" : "pass",
|
|
433
|
+
success: failures === 0,
|
|
434
|
+
error: failures > 0 ? `${failures} of ${runs.length} test case${runs.length === 1 ? "" : "s"} failed` : null,
|
|
435
|
+
completedAt: new Date().toISOString(),
|
|
436
|
+
durationMs: Math.max(aggregateStats.durationMs, Date.now() - batchStartedAt),
|
|
437
|
+
platform: deviceInfo.platform,
|
|
438
|
+
deviceSource: deviceInfo.source,
|
|
439
|
+
deviceName: deviceInfo.device_name,
|
|
440
|
+
osVersion: deviceInfo.os_version,
|
|
441
|
+
instructionsTotal: aggregateStats.instructionsTotal,
|
|
442
|
+
instructionsCompleted: aggregateStats.instructionsCompleted,
|
|
443
|
+
actionsTotal: aggregateStats.actionsTotal,
|
|
444
|
+
assertionsPassed: aggregateStats.assertionsPassed,
|
|
445
|
+
assertionsFailed: aggregateStats.assertionsFailed,
|
|
446
|
+
retries: aggregateStats.retries,
|
|
447
|
+
onWarning: (message) => console.warn(`Warning: ${message}`),
|
|
448
|
+
});
|
|
449
|
+
}
|
|
288
450
|
return {
|
|
289
451
|
exitCode: failures === 0 ? 0 : 1,
|
|
290
452
|
failures
|
|
@@ -32,7 +32,7 @@ export async function handleHelp(args, session, context) {
|
|
|
32
32
|
addOutput({ type: 'info', text: ' --session-mode <mode> Browser session mode for web runs: persistent or fresh' });
|
|
33
33
|
addOutput({ type: 'info', text: ' --config <file> Headless execution JSON config for prompt parity' });
|
|
34
34
|
addOutput({ type: 'info', text: ' --llm-provider <mode> Headless AI provider: loadmill or openai' });
|
|
35
|
-
addOutput({ type: 'info', text: ' --cua-model <model> Headless CUA model override: gpt-5.4, loadmill-smart, or gpt-5.4-mini' });
|
|
35
|
+
addOutput({ type: 'info', text: ' --cua-model <model> Headless CUA model override: gpt-5.4, loadmill-smart, loadmill-beacon, or gpt-5.4-mini' });
|
|
36
36
|
addOutput({ type: 'info', text: ' --aim-assist <mode> Headless Aim Assist mode: off, smart, or always' });
|
|
37
37
|
addOutput({ type: 'info', text: ' --context <file> Optional app context file injected into execution runs' });
|
|
38
38
|
addOutput({ type: 'info', text: ' --no-context Disable app context for headless execution' });
|
|
@@ -2,15 +2,22 @@ export const DEFAULT_OPENAI_CUA_MODEL = "gpt-5.4";
|
|
|
2
2
|
export const DEFAULT_LOADMILL_CUA_MODEL = "loadmill-smart";
|
|
3
3
|
export const DEFAULT_CUA_MODEL = DEFAULT_LOADMILL_CUA_MODEL;
|
|
4
4
|
export const OPENAI_CUA_MODELS = ["gpt-5.4", "gpt-5.4-mini"];
|
|
5
|
-
export const LOADMILL_CUA_MODELS = [DEFAULT_LOADMILL_CUA_MODEL];
|
|
6
|
-
export const CUA_MODELS = ["gpt-5.4", "loadmill-smart", "gpt-5.4-mini"];
|
|
5
|
+
export const LOADMILL_CUA_MODELS = [DEFAULT_LOADMILL_CUA_MODEL, "loadmill-beacon"];
|
|
6
|
+
export const CUA_MODELS = ["gpt-5.4", "loadmill-smart", "loadmill-beacon", "gpt-5.4-mini"];
|
|
7
7
|
export const CUA_MODEL_OPTIONS = Object.freeze([
|
|
8
8
|
Object.freeze({
|
|
9
9
|
id: "loadmill-smart",
|
|
10
10
|
providerMode: "loadmill",
|
|
11
11
|
providerLabel: "Loadmill",
|
|
12
|
-
name: "Loadmill Smart",
|
|
13
|
-
label: "Loadmill Smart (recommended)",
|
|
12
|
+
name: "Loadmill - Smart",
|
|
13
|
+
label: "Loadmill - Smart (recommended)",
|
|
14
|
+
}),
|
|
15
|
+
Object.freeze({
|
|
16
|
+
id: "loadmill-beacon",
|
|
17
|
+
providerMode: "loadmill",
|
|
18
|
+
providerLabel: "Loadmill",
|
|
19
|
+
name: "Loadmill - Beacon",
|
|
20
|
+
label: "Loadmill - Beacon",
|
|
14
21
|
}),
|
|
15
22
|
Object.freeze({
|
|
16
23
|
id: "gpt-5.4",
|
|
@@ -0,0 +1,106 @@
|
|
|
1
|
+
const TOKENS_PER_MILLION = 1_000_000;
|
|
2
|
+
const OPENAI_GPT_5_4_PRICING = Object.freeze({
|
|
3
|
+
inputPerMillionUsd: 2.5,
|
|
4
|
+
});
|
|
5
|
+
const OPENAI_GPT_5_4_MINI_PRICING = Object.freeze({
|
|
6
|
+
inputPerMillionUsd: 0.75,
|
|
7
|
+
});
|
|
8
|
+
const MODEL_PRICING = Object.freeze({
|
|
9
|
+
"gpt-5.4": Object.freeze({
|
|
10
|
+
...OPENAI_GPT_5_4_PRICING,
|
|
11
|
+
}),
|
|
12
|
+
"gpt-5.4-mini": Object.freeze({
|
|
13
|
+
...OPENAI_GPT_5_4_MINI_PRICING,
|
|
14
|
+
}),
|
|
15
|
+
// Loadmill Smart currently routes to GPT-5.4 behind the Loadmill LLM gateway.
|
|
16
|
+
// Keep this explicit so pricing can be updated here if the backing model changes.
|
|
17
|
+
"loadmill-smart": Object.freeze({
|
|
18
|
+
...OPENAI_GPT_5_4_PRICING,
|
|
19
|
+
}),
|
|
20
|
+
});
|
|
21
|
+
function finiteNumber(value) {
|
|
22
|
+
return typeof value === "number" && Number.isFinite(value) && value > 0 ? value : 0;
|
|
23
|
+
}
|
|
24
|
+
function trimToNull(value) {
|
|
25
|
+
if (typeof value !== "string") {
|
|
26
|
+
return null;
|
|
27
|
+
}
|
|
28
|
+
const trimmed = value.trim();
|
|
29
|
+
return trimmed || null;
|
|
30
|
+
}
|
|
31
|
+
export function normalizeDroidCuaCostModel(model) {
|
|
32
|
+
if (typeof model !== "string") {
|
|
33
|
+
return null;
|
|
34
|
+
}
|
|
35
|
+
if (model === "gpt-5.4-mini" || model.startsWith("gpt-5.4-mini-")) {
|
|
36
|
+
return "gpt-5.4-mini";
|
|
37
|
+
}
|
|
38
|
+
if (model === "gpt-5.4" || model.startsWith("gpt-5.4-")) {
|
|
39
|
+
return "gpt-5.4";
|
|
40
|
+
}
|
|
41
|
+
if (model === "loadmill-smart" || model === "smart") {
|
|
42
|
+
return "loadmill-smart";
|
|
43
|
+
}
|
|
44
|
+
return null;
|
|
45
|
+
}
|
|
46
|
+
export function createDroidCuaRunUsageTotals({ model = null } = {}) {
|
|
47
|
+
const normalizedModel = trimToNull(model);
|
|
48
|
+
return {
|
|
49
|
+
model: normalizedModel,
|
|
50
|
+
costModel: normalizeDroidCuaCostModel(normalizedModel),
|
|
51
|
+
inputTokens: 0,
|
|
52
|
+
cachedInputTokens: 0,
|
|
53
|
+
outputTokens: 0,
|
|
54
|
+
reasoningOutputTokens: 0,
|
|
55
|
+
totalTokens: 0,
|
|
56
|
+
};
|
|
57
|
+
}
|
|
58
|
+
export function setDroidCuaRunUsageModel(totals, model) {
|
|
59
|
+
const normalizedModel = trimToNull(model);
|
|
60
|
+
if (!normalizedModel) {
|
|
61
|
+
return totals;
|
|
62
|
+
}
|
|
63
|
+
return {
|
|
64
|
+
...totals,
|
|
65
|
+
model: totals.model ?? normalizedModel,
|
|
66
|
+
costModel: totals.costModel ?? normalizeDroidCuaCostModel(normalizedModel),
|
|
67
|
+
};
|
|
68
|
+
}
|
|
69
|
+
export function addOpenAiResponseToDroidCuaRunUsage(totals, response) {
|
|
70
|
+
const record = response && typeof response === "object" ? response : {};
|
|
71
|
+
const usage = record.usage && typeof record.usage === "object" ? record.usage : null;
|
|
72
|
+
if (!usage) {
|
|
73
|
+
return totals;
|
|
74
|
+
}
|
|
75
|
+
const inputDetails = usage.input_tokens_details && typeof usage.input_tokens_details === "object"
|
|
76
|
+
? usage.input_tokens_details
|
|
77
|
+
: {};
|
|
78
|
+
const outputDetails = usage.output_tokens_details && typeof usage.output_tokens_details === "object"
|
|
79
|
+
? usage.output_tokens_details
|
|
80
|
+
: {};
|
|
81
|
+
const observedModel = trimToNull(record.model);
|
|
82
|
+
const inputTokens = finiteNumber(usage.input_tokens);
|
|
83
|
+
const outputTokens = finiteNumber(usage.output_tokens);
|
|
84
|
+
const totalTokens = finiteNumber(usage.total_tokens) || inputTokens + outputTokens;
|
|
85
|
+
return {
|
|
86
|
+
model: totals.model ?? observedModel,
|
|
87
|
+
costModel: totals.costModel ?? normalizeDroidCuaCostModel(observedModel),
|
|
88
|
+
inputTokens: totals.inputTokens + inputTokens,
|
|
89
|
+
cachedInputTokens: totals.cachedInputTokens + finiteNumber(inputDetails.cached_tokens),
|
|
90
|
+
outputTokens: totals.outputTokens + outputTokens,
|
|
91
|
+
reasoningOutputTokens: totals.reasoningOutputTokens + finiteNumber(outputDetails.reasoning_tokens),
|
|
92
|
+
totalTokens: totals.totalTokens + totalTokens,
|
|
93
|
+
};
|
|
94
|
+
}
|
|
95
|
+
export function estimateDroidCuaRunCostUsd(totals) {
|
|
96
|
+
const costModel = totals.costModel ?? normalizeDroidCuaCostModel(totals.model);
|
|
97
|
+
if (!costModel) {
|
|
98
|
+
return null;
|
|
99
|
+
}
|
|
100
|
+
const pricing = MODEL_PRICING[costModel];
|
|
101
|
+
if (!pricing) {
|
|
102
|
+
return null;
|
|
103
|
+
}
|
|
104
|
+
const cost = (totals.inputTokens / TOKENS_PER_MILLION) * pricing.inputPerMillionUsd;
|
|
105
|
+
return Number(cost.toFixed(8));
|
|
106
|
+
}
|
|
@@ -0,0 +1,32 @@
|
|
|
1
|
+
export const DroidCuaSegmentEvent = Object.freeze({
|
|
2
|
+
TEST_RUN_COMPLETED: "Droid CUA test run completed",
|
|
3
|
+
});
|
|
4
|
+
const VALID_TEST_RUN_STATUSES = new Set(["pass", "failure", "error", "skipped"]);
|
|
5
|
+
function finiteNumberOrNull(value) {
|
|
6
|
+
return typeof value === "number" && Number.isFinite(value) ? value : null;
|
|
7
|
+
}
|
|
8
|
+
export function normalizeDroidCuaTestRunStatus(value) {
|
|
9
|
+
return VALID_TEST_RUN_STATUSES.has(value) ? value : "error";
|
|
10
|
+
}
|
|
11
|
+
export function normalizeDroidCuaTestRunDevice(platform) {
|
|
12
|
+
if (platform === "ios")
|
|
13
|
+
return "ios";
|
|
14
|
+
if (platform === "android")
|
|
15
|
+
return "android";
|
|
16
|
+
return "web";
|
|
17
|
+
}
|
|
18
|
+
export function buildDroidCuaTestRunCompletedProperties(input = {}) {
|
|
19
|
+
const estimatedCostUsd = finiteNumberOrNull(input.estimatedCostUsd);
|
|
20
|
+
const tokensUsed = finiteNumberOrNull(input.tokensUsed);
|
|
21
|
+
return {
|
|
22
|
+
status: normalizeDroidCuaTestRunStatus(input.status),
|
|
23
|
+
durationMs: Math.max(0, Math.floor(finiteNumberOrNull(input.durationMs) ?? 0)),
|
|
24
|
+
numOfSteps: Math.max(0, Math.floor(finiteNumberOrNull(input.numOfSteps) ?? 0)),
|
|
25
|
+
device: normalizeDroidCuaTestRunDevice(input.platform),
|
|
26
|
+
...(input.teamId ? { teamId: input.teamId } : {}),
|
|
27
|
+
...(input.model ? { model: input.model } : {}),
|
|
28
|
+
...(input.llmProviderMode ? { llmProviderMode: input.llmProviderMode } : {}),
|
|
29
|
+
...(estimatedCostUsd != null ? { estimatedCostUsd } : {}),
|
|
30
|
+
...(tokensUsed != null ? { tokensUsed } : {}),
|
|
31
|
+
};
|
|
32
|
+
}
|