@loadmill/droid-cua 2.8.0 → 2.9.0

package/build/src/integrations/loadmill/client.js

@@ -20,153 +20,171 @@ function normalizeApiBaseUrl(rawBaseUrl) {
     }
 }
 /**
- * Get Loadmill API token from environment
- * @returns {string|null}
+ * @param {{
+ *   token?: string | null,
+ *   baseUrl?: string | null,
+ *   onUnauthorized?: (() => Promise<unknown> | unknown) | null,
+ * }} [options]
  */
-export function getApiToken() {
-    return process.env.LOADMILL_API_TOKEN || null;
-}
-/**
- * Get Loadmill base URL from environment
- * @returns {string}
- */
-export function getBaseUrl() {
-    return normalizeApiBaseUrl(process.env.LOADMILL_BASE_URL);
-}
-/**
- * Make an authenticated request to Loadmill API
- * @param {string} endpoint - API endpoint (without base URL)
- * @param {Object} options - Fetch options
- * @returns {Promise<Object>}
- */
-async function apiRequest(endpoint, options = {}) {
-    const token = getApiToken();
-    if (!token) {
-        throw new Error("LOADMILL_API_TOKEN environment variable is not set");
+export function createLoadmillApiClient({ token = null, baseUrl = null, onUnauthorized = null, } = {}) {
+    function getApiToken() {
+        return typeof token === "string" && token.trim() ? token.trim() : null;
+    }
+    function getBaseUrl() {
+        return normalizeApiBaseUrl(baseUrl);
     }
-    const baseUrl = getBaseUrl();
-    const url = `${baseUrl}${endpoint}`;
-    const response = await fetch(url, {
-        ...options,
-        headers: {
-            "Authorization": `Bearer ${token}`,
-            "Content-Type": "application/json",
-            ...options.headers,
-        },
-    });
-    if (response.status === 401 || response.status === 403) {
-        const unauthorizedHandler = globalThis.__DROID_CUA_HANDLE_LOADMILL_UNAUTHORIZED__;
-        if (typeof unauthorizedHandler === "function") {
-            await unauthorizedHandler();
+    async function apiRequest(endpoint, options = {}) {
+        const resolvedToken = getApiToken();
+        if (!resolvedToken) {
+            throw new Error("LOADMILL_API_TOKEN environment variable is not set");
+        }
+        const resolvedBaseUrl = getBaseUrl();
+        const url = `${resolvedBaseUrl}${endpoint}`;
+        const response = await fetch(url, {
+            ...options,
+            headers: {
+                "Authorization": `Bearer ${resolvedToken}`,
+                "Content-Type": "application/json",
+                ...options.headers,
+            },
+        });
+        if ((response.status === 401 || response.status === 403) && typeof onUnauthorized === "function") {
+            await onUnauthorized();
+        }
+        if (!response.ok) {
+            const errorText = await response.text();
+            throw new Error(`Loadmill API error (${response.status}): ${errorText}`);
         }
+        return response.json();
     }
-    if (!response.ok) {
-        const errorText = await response.text();
-        throw new Error(`Loadmill API error (${response.status}): ${errorText}`);
+    async function searchFlows(searchQuery, options = {}) {
+        const { limit = 10 } = options;
+        const encodedQuery = encodeURIComponent(searchQuery);
+        const result = await apiRequest(`/test-flows?search=${encodedQuery}&limit=${limit}`);
+        if (Array.isArray(result)) {
+            return result;
+        }
+        if (result && Array.isArray(result.flows)) {
+            return result.flows;
+        }
+        if (result && Array.isArray(result.data)) {
+            return result.data;
+        }
+        if (result && Array.isArray(result.items)) {
+            return result.items;
+        }
+        if (result && Array.isArray(result.testFlows)) {
+            return result.testFlows;
+        }
+        console.error("[Loadmill] Unexpected API response format:", JSON.stringify(result, null, 2));
+        return [];
     }
-    return response.json();
-}
-/**
- * Search for test flows by query
- * @param {string} searchQuery - Search query string
- * @param {Object} options - Additional options
- * @param {number} options.limit - Maximum number of results (default: 10)
- * @returns {Promise<Array>} - Array of matching flows
- */
-export async function searchFlows(searchQuery, options = {}) {
-    const { limit = 10 } = options;
-    const encodedQuery = encodeURIComponent(searchQuery);
-    const result = await apiRequest(`/test-flows?search=${encodedQuery}&limit=${limit}`);
-    // Handle different response formats
-    if (Array.isArray(result)) {
-        return result;
+    async function runTestSuite(suiteId, { flowIds = [], parameters = {} } = {}) {
+        const body = {};
+        if (flowIds.length > 0) {
+            body.flowIds = flowIds;
+        }
+        if (Object.keys(parameters).length > 0) {
+            body.parameters = parameters;
+        }
+        return await apiRequest(`/test-suites/${suiteId}/run`, {
+            method: "POST",
+            body: JSON.stringify(body),
+        });
     }
-    if (result && Array.isArray(result.flows)) {
-        return result.flows;
+    async function runTestFlow(flowId, suiteId, { parameters = {} } = {}) {
+        const body = {
+            flows: [flowId],
+            inlineParameterOverride: true,
+            sharedFlowVersionOverrides: [],
+        };
+        if (Object.keys(parameters).length > 0) {
+            body.overrideParameters = parameters;
+        }
+        return await apiRequest(`/test-suites/${suiteId}/run?ui=true`, {
+            method: "POST",
+            body: JSON.stringify(body),
+        });
     }
-    if (result && Array.isArray(result.data)) {
-        return result.data;
+    async function getTestRunStatus(testRunId) {
+        return await apiRequest(`/test-suites-runs/${testRunId}`);
     }
-    if (result && Array.isArray(result.items)) {
-        return result.items;
+    async function getFlowRunStatus(flowRunId) {
+        return await apiRequest(`/test-flows-runs/${flowRunId}`);
     }
-    if (result && Array.isArray(result.testFlows)) {
-        return result.testFlows;
+    async function explainFailure(testRunId) {
+        return await apiRequest(`/explain-failures`, {
+            method: "POST",
+            body: JSON.stringify({
+                testRunId,
+                testRunType: "flowRun"
+            }),
+        });
     }
-    // Log unexpected response format for debugging
-    console.error("[Loadmill] Unexpected API response format:", JSON.stringify(result, null, 2));
-    return [];
+    return {
+        getApiToken,
+        getBaseUrl,
+        searchFlows,
+        runTestSuite,
+        runTestFlow,
+        getTestRunStatus,
+        getFlowRunStatus,
+        explainFailure,
+    };
 }
-/**
- * Run a test suite with specified flows
- * @param {string} suiteId - Test suite ID
- * @param {Object} options - Run options
- * @param {string[]} options.flowIds - Array of flow IDs to run
- * @param {Object} options.parameters - Parameters to pass to the flows
- * @returns {Promise<Object>} - Run result with runId
- */
-export async function runTestSuite(suiteId, { flowIds = [], parameters = {} } = {}) {
-    const body = {};
-    if (flowIds.length > 0) {
-        body.flowIds = flowIds;
-    }
-    if (Object.keys(parameters).length > 0) {
-        body.parameters = parameters;
-    }
-    return apiRequest(`/test-suites/${suiteId}/run`, {
-        method: "POST",
-        body: JSON.stringify(body),
-    });
+export function getApiToken() {
+    return createLoadmillApiClient({
+        token: process.env.LOADMILL_API_TOKEN,
+        baseUrl: process.env.LOADMILL_BASE_URL,
+    }).getApiToken();
 }
-/**
- * Run a single test flow
- * @param {string} flowId - Flow ID to run
- * @param {string} suiteId - Test suite ID containing the flow
- * @param {Object} options - Run options
- * @param {Object} options.parameters - Parameters to pass to the flow
- * @returns {Promise<Object>} - Run result with testSuiteRunId
- */
-export async function runTestFlow(flowId, suiteId, { parameters = {} } = {}) {
-    const body = {
-        flows: [flowId],
-        inlineParameterOverride: true,
-        sharedFlowVersionOverrides: [],
-    };
-    if (Object.keys(parameters).length > 0) {
-        body.overrideParameters = parameters;
-    }
-    return apiRequest(`/test-suites/${suiteId}/run?ui=true`, {
-        method: "POST",
-        body: JSON.stringify(body),
-    });
+export function getBaseUrl() {
+    return createLoadmillApiClient({
+        token: process.env.LOADMILL_API_TOKEN,
+        baseUrl: process.env.LOADMILL_BASE_URL,
+    }).getBaseUrl();
+}
+export async function searchFlows(searchQuery, options = {}) {
+    // Legacy env-based wrapper kept for older call sites. New execution flows
+    // should pass an explicit client created from resolved run credentials.
+    return await createLoadmillApiClient({
+        token: process.env.LOADMILL_API_TOKEN,
+        baseUrl: process.env.LOADMILL_BASE_URL,
+        onUnauthorized: globalThis.__DROID_CUA_HANDLE_LOADMILL_UNAUTHORIZED__,
+    }).searchFlows(searchQuery, options);
+}
+export async function runTestSuite(suiteId, options = {}) {
+    return await createLoadmillApiClient({
+        token: process.env.LOADMILL_API_TOKEN,
+        baseUrl: process.env.LOADMILL_BASE_URL,
+        onUnauthorized: globalThis.__DROID_CUA_HANDLE_LOADMILL_UNAUTHORIZED__,
+    }).runTestSuite(suiteId, options);
+}
+export async function runTestFlow(flowId, suiteId, options = {}) {
+    return await createLoadmillApiClient({
+        token: process.env.LOADMILL_API_TOKEN,
+        baseUrl: process.env.LOADMILL_BASE_URL,
+        onUnauthorized: globalThis.__DROID_CUA_HANDLE_LOADMILL_UNAUTHORIZED__,
+    }).runTestFlow(flowId, suiteId, options);
 }
-/**
- * Get the status of a test run
- * @param {string} testRunId - Test run ID
- * @returns {Promise<Object>} - Test run status
- */
 export async function getTestRunStatus(testRunId) {
-    return apiRequest(`/test-suites-runs/${testRunId}`);
+    return await createLoadmillApiClient({
+        token: process.env.LOADMILL_API_TOKEN,
+        baseUrl: process.env.LOADMILL_BASE_URL,
+        onUnauthorized: globalThis.__DROID_CUA_HANDLE_LOADMILL_UNAUTHORIZED__,
+    }).getTestRunStatus(testRunId);
 }
-/**
- * Get the status of a flow run
- * @param {string} flowRunId - Flow run ID
- * @returns {Promise<Object>} - Flow run status
- */
 export async function getFlowRunStatus(flowRunId) {
-    return apiRequest(`/test-flows-runs/${flowRunId}`);
+    return await createLoadmillApiClient({
+        token: process.env.LOADMILL_API_TOKEN,
+        baseUrl: process.env.LOADMILL_BASE_URL,
+        onUnauthorized: globalThis.__DROID_CUA_HANDLE_LOADMILL_UNAUTHORIZED__,
+    }).getFlowRunStatus(flowRunId);
 }
-/**
- * Get AI-generated explanation for a failed test run
- * @param {string} testRunId - The flow run ID (from testSuiteFlowRuns[].id)
- * @returns {Promise<Object>} - Explanation with reason, suggestion, etc.
- */
 export async function explainFailure(testRunId) {
-    return apiRequest(`/explain-failures`, {
-        method: "POST",
-        body: JSON.stringify({
-            testRunId,
-            testRunType: "flowRun"
-        }),
-    });
+    return await createLoadmillApiClient({
+        token: process.env.LOADMILL_API_TOKEN,
+        baseUrl: process.env.LOADMILL_BASE_URL,
+        onUnauthorized: globalThis.__DROID_CUA_HANDLE_LOADMILL_UNAUTHORIZED__,
+    }).explainFailure(testRunId);
 }

package/build/src/integrations/loadmill/executor.js

@@ -1,7 +1,7 @@
 /**
  * Orchestrates Loadmill flow execution
  */
-import { searchFlows, runTestFlow, getTestRunStatus, getApiToken, explainFailure } from "./client.js";
+import { createLoadmillApiClient } from "./client.js";
 import { interpretLoadmillCommand, selectBestFlow } from "./interpreter.js";
 import { logger } from "../../utils/logger.js";
 const POLL_INTERVAL_MS = 5000; // 5 seconds
@@ -20,11 +20,11 @@ function sleep(ms) {
  * @param {Function} onStatusUpdate - Callback for status updates
  * @returns {Promise<Object>} - Final run status
  */
-async function pollForCompletion(runId, onStatusUpdate = () => { }) {
+async function pollForCompletion(runId, loadmillClient, onStatusUpdate = () => { }) {
     const startTime = Date.now();
     while (Date.now() - startTime < TIMEOUT_MS) {
         try {
-            const status = await getTestRunStatus(runId);
+            const status = await loadmillClient.getTestRunStatus(runId);
             onStatusUpdate(status);
             // Check if run is complete
             if (status.status === "PASSED" || status.status === "passed") {
@@ -36,7 +36,7 @@ async function pollForCompletion(runId, onStatusUpdate = () => { }) {
                 try {
                     const failedFlowRun = status.testSuiteFlowRuns?.find(f => f.status === "FAILED");
                     if (failedFlowRun) {
-                        const explanation = await explainFailure(failedFlowRun.id);
+                        const explanation = await loadmillClient.explainFailure(failedFlowRun.id);
                         if (explanation?.reason) {
                             failureReason = explanation.reason;
                         }
@@ -69,22 +69,28 @@ async function pollForCompletion(runId, onStatusUpdate = () => { }) {
  * @returns {Promise<{success: boolean, error?: string, result?: Object}>}
  */
 export async function executeLoadmillCommand(userInput, options = {}) {
-    const { onProgress = () => { } } = options;
+    const { onProgress = () => { }, loadmillClient = createLoadmillApiClient(), openAiClient = null, } = options;
     // Check for API token
-    if (!getApiToken()) {
+    if (!loadmillClient.getApiToken()) {
         return {
             success: false,
             error: "LOADMILL_API_TOKEN environment variable is not set. Please set it in your .env file."
         };
     }
+    if (!openAiClient) {
+        return {
+            success: false,
+            error: "OpenAI API key is not configured."
+        };
+    }
     try {
         // Step 1: Interpret the command
         onProgress({ step: "interpreting", message: "Interpreting command..." });
-        const interpreted = await interpretLoadmillCommand(userInput);
+        const interpreted = await interpretLoadmillCommand(userInput, { client: openAiClient });
         logger.debug("Loadmill command interpreted", interpreted);
         // Step 2: Search for flows
         onProgress({ step: "searching", message: `Searching for flows matching "${interpreted.searchQuery}"...` });
-        const flows = await searchFlows(interpreted.searchQuery);
+        const flows = await loadmillClient.searchFlows(interpreted.searchQuery);
         // Ensure flows is an array
         if (!Array.isArray(flows) || flows.length === 0) {
             return {
@@ -94,7 +100,7 @@ export async function executeLoadmillCommand(userInput, options = {}) {
         }
         // Step 3: Select best match
         onProgress({ step: "selecting", message: `Found ${flows.length} flow(s). Selecting best match...` });
-        const { selectedFlow, confidence } = await selectBestFlow(flows, interpreted.searchQuery);
+        const { selectedFlow, confidence } = await selectBestFlow(flows, interpreted.searchQuery, { client: openAiClient });
         if (!selectedFlow) {
             return {
                 success: false,
@@ -119,7 +125,7 @@ export async function executeLoadmillCommand(userInput, options = {}) {
             step: "running",
             message: `Running flow "${flowName}" (confidence: ${(confidence * 100).toFixed(0)}%)...`
         });
-        const runResult = await runTestFlow(selectedFlow.id, selectedFlow.testSuiteId, {
+        const runResult = await loadmillClient.runTestFlow(selectedFlow.id, selectedFlow.testSuiteId, {
             parameters: interpreted.parameters
         });
         const runId = runResult.testSuiteRunId || runResult.id || runResult.runId;
@@ -131,7 +137,7 @@ export async function executeLoadmillCommand(userInput, options = {}) {
         }
         // Step 5: Poll for completion
         onProgress({ step: "polling", message: `Test started (ID: ${runId}). Waiting for completion...`, runId });
-        const finalResult = await pollForCompletion(runId, (status) => {
+        const finalResult = await pollForCompletion(runId, loadmillClient, (status) => {
             onProgress({ step: "polling", message: `Status: ${status.status}...`, runId });
         });
         return {

package/build/src/integrations/loadmill/index.js

@@ -1,6 +1,6 @@
 /**
  * Loadmill integration public exports
  */
-export { getApiToken, searchFlows, runTestFlow, getFlowRunStatus } from "./client.js";
+export { createLoadmillApiClient, getApiToken, searchFlows, runTestFlow, getFlowRunStatus } from "./client.js";
 export { interpretLoadmillCommand, selectBestFlow } from "./interpreter.js";
 export { executeLoadmillCommand } from "./executor.js";

package/build/src/integrations/loadmill/interpreter.js

@@ -1,24 +1,14 @@
 /**
  * AI-powered text interpretation for Loadmill commands
  */
-import OpenAI from "openai";
 import { buildLoadmillCommandInterpretationMessages, buildLoadmillFlowSelectionMessages, } from "../../prompts/loadmill.js";
-let openai = null;
-function getOpenAI() {
-    if (!openai) {
-        openai = new OpenAI({
-            apiKey: process.env.OPENAI_API_KEY,
-        });
-    }
-    return openai;
-}
 /**
  * Interpret a natural language Loadmill command into structured data
  * @param {string} userInput - Natural language command
  * @returns {Promise<{searchQuery: string, parameters: Object, action: 'run'|'search'}>}
  */
-export async function interpretLoadmillCommand(userInput) {
-    const response = await getOpenAI().chat.completions.create({
+export async function interpretLoadmillCommand(userInput, { client }) {
+    const response = await client.chat.completions.create({
         model: "gpt-4o-mini",
         messages: buildLoadmillCommandInterpretationMessages(userInput),
         response_format: { type: "json_object" }
@@ -37,7 +27,7 @@ export async function interpretLoadmillCommand(userInput) {
  * @param {string} originalQuery - Original user query
  * @returns {Promise<{selectedFlow: Object|null, confidence: number}>}
  */
-export async function selectBestFlow(flows, originalQuery) {
+export async function selectBestFlow(flows, originalQuery, { client }) {
     // Ensure flows is an array
     if (!flows || !Array.isArray(flows) || flows.length === 0) {
         return { selectedFlow: null, confidence: 0 };
@@ -51,7 +41,7 @@ export async function selectBestFlow(flows, originalQuery) {
         const suite = f.testSuiteDescription || "";
         return `${i + 1}. ID: ${f.id}, Name: "${name}"${suite ? `, Suite: "${suite}"` : ""}`;
     }).join("\n");
-    const response = await getOpenAI().chat.completions.create({
+    const response = await client.chat.completions.create({
         model: "gpt-4o-mini",
         messages: buildLoadmillFlowSelectionMessages(originalQuery, flowList),
         response_format: { type: "json_object" }

package/build/src/modes/design-mode-ink.js

@@ -1,5 +1,4 @@
 import { getScreenshotAsBase64 } from "../device/connection.js";
-import { sendCUARequest, reviseTestScript } from "../device/openai.js";
 import { buildDesignModePrompt, buildDesignRecoveryPrompt } from "../core/prompts.js";
 import { saveTest } from "../test-store/test-manager.js";
 import { logger } from "../utils/logger.js";
@@ -23,6 +22,8 @@ export class DesignModeInk {
         this.baseDesignPrompt = null;
         this.consecutiveErrorCount = 0;
         this.maxConsecutiveErrors = 3;
+        this.sendCUARequest = executionEngine.sendCUARequest;
+        this.reviseTestScript = executionEngine.reviseTestScript;
     }
     /**
      * Start design mode conversation
@@ -183,7 +184,7 @@ export class DesignModeInk {
                 }
                 // Get screenshot and send to model
                 const screenshotBase64 = await getScreenshotAsBase64(this.session.deviceId, this.session.deviceInfo);
-                const response = await sendCUARequest({
+                const response = await this.sendCUARequest({
                     messages: this.session.messages,
                     screenshotBase64,
                     previousResponseId: this.session.previousResponseId,
@@ -271,7 +272,7 @@ export class DesignModeInk {
                             const revision = confirm;
                             // Use simple chat completion for revision instead of CUA
                             addOutput({ type: 'info', text: 'Revising test script...' });
-                            currentScript = await reviseTestScript(currentScript, revision);
+                            currentScript = await this.reviseTestScript(currentScript, revision);
                             // Display the revised script
                             addOutput({ type: 'info', text: '' });
                             addOutput({ type: 'system', text: '=== Generated Test Script ===' });

package/build/src/modes/design-mode.js

@@ -1,6 +1,5 @@
 import readline from "readline";
 import { getScreenshotAsBase64 } from "../device/connection.js";
-import { sendCUARequest, reviseTestScript } from "../device/openai.js";
 import { buildDesignModePrompt, buildDesignRecoveryPrompt } from "../core/prompts.js";
 import { saveTest } from "../test-store/test-manager.js";
 import { logger } from "../utils/logger.js";
@@ -20,6 +19,8 @@ export class DesignMode {
         this.baseDesignPrompt = null;
         this.consecutiveErrorCount = 0;
         this.maxConsecutiveErrors = 3;
+        this.sendCUARequest = executionEngine.sendCUARequest;
+        this.reviseTestScript = executionEngine.reviseTestScript;
     }
     /**
      * Start design mode conversation
@@ -147,7 +148,7 @@ export class DesignMode {
                 const keypressHandler = this.setupEscDetection();
                 // Get screenshot and send to model
                 const screenshotBase64 = await getScreenshotAsBase64(this.session.deviceId, this.session.deviceInfo);
-                const response = await sendCUARequest({
+                const response = await this.sendCUARequest({
                     messages: this.session.messages,
                     screenshotBase64,
                     previousResponseId: this.session.previousResponseId,
@@ -262,7 +263,7 @@ export class DesignMode {
                             const revision = confirm;
                             // Use simple chat completion for revision instead of CUA
                             console.log("Revising test script...");
-                            currentScript = await reviseTestScript(currentScript, revision);
+                            currentScript = await this.reviseTestScript(currentScript, revision);
                             // Display the revised script
                             console.log("\n=== Generated Test Script ===");
                             console.log(currentScript);

package/build/src/modes/execution-mode.js

@@ -228,6 +228,14 @@ export class ExecutionMode {
             }
             const navigationTarget = extractNavigateTarget(instruction);
             this.session.addToTranscript(`[Navigate] ${navigationTarget}`);
+            emitDesktopDebug("runner.navigate.started", "execution", {
+                runId: context?.runId,
+                stepId: stepContext?.stepId,
+                instructionIndex: stepContext?.instructionIndex
+            }, {
+                instruction,
+                target: navigationTarget
+            });
             this.emit(addOutput, "system", `Navigating to ${navigationTarget}...`, context, stepContext, {
                 eventType: "system_message",
                 payload: {
@@ -250,6 +258,15 @@ export class ExecutionMode {
                     navigatedUrl: this.session.currentUrl,
                 });
                 this.session.installRecoverySystemPrompt(navigationRecoveryPrompt);
+                emitDesktopDebug("runner.navigate.completed", "execution", {
+                    runId: context?.runId,
+                    stepId: stepContext?.stepId,
+                    instructionIndex: stepContext?.instructionIndex
+                }, {
+                    instruction,
+                    target: navigationTarget,
+                    url: normalizedUrl
+                });
                 this.emit(addOutput, "success", `Navigation complete: ${normalizedUrl}`, context, stepContext, {
                     eventType: "system_message",
                     payload: {
@@ -266,6 +283,15 @@ export class ExecutionMode {
                     retryCount,
                     message,
                 });
+                emitDesktopDebug("runner.navigate.failed", "execution", {
+                    runId: context?.runId,
+                    stepId: stepContext?.stepId,
+                    instructionIndex: stepContext?.instructionIndex
+                }, {
+                    instruction,
+                    target: navigationTarget,
+                    message
+                });
                 this.emit(addOutput, "error", `Navigation failed: ${message}`, context, stepContext, {
                     eventType: "error",
                     payload: {