testdriverai 7.2.64 → 7.2.65

package/lib/sentry.js

@@ -14,6 +14,7 @@ const Sentry = require("@sentry/node");
 const crypto = require("crypto");
 const os = require("os");
 const { version } = require("../package.json");
+const logger = require("../agent/lib/logger");
 
 // Store the current session's trace context
 let currentTraceId = null;
@@ -31,7 +32,7 @@ const isEnabled = () => {
 };
 
 if (isEnabled()) {
-  console.log("Analytics enabled. Set TD_TELEMETRY=false to disable.");
+  logger.log("Analytics enabled. Set TD_TELEMETRY=false to disable.");
   Sentry.init({
     dsn:
       process.env.SENTRY_DSN ||
@@ -174,7 +175,7 @@ function setSessionTraceContext(sessionId) {
     }
   } catch (e) {
     // Ignore errors - propagation context may not be supported
-    console.log("Could not set propagation context:", e.message);
+    logger.log("Could not set propagation context:", e.message);
   }
 }
 
@@ -203,7 +204,7 @@ function attachLogListeners(emitter) {
 
   // Check if Sentry.logger is available
   if (!Sentry.logger) {
-    console.log("Sentry.logger not available, skipping log listeners");
+    logger.log("Sentry.logger not available, skipping log listeners");
     return;
   }
 

package/lib/vitest/hooks.mjs

@@ -248,6 +248,7 @@ export function TestDriver(context, options = {}) {
 
   const testdriver = new TestDriverSDK(apiKey, config);
   testdriver.__vitestContext = context.task;
+  testdriver._debugOnFailure = mergedOptions.debugOnFailure || false;
   testDriverInstances.set(context.task, testdriver);
 
   // Set platform metadata early so the reporter can show the correct OS from the start
@@ -298,29 +299,29 @@ export function TestDriver(context, options = {}) {
       );
     }
 
-    // Set up console spy using vi.spyOn (test-isolated)
-    setupConsoleSpy(testdriver, context.task.id);
+    // Only set up dashcam-related infrastructure if enabled (default: true)
+    if (testdriver.dashcamEnabled) {
+      // Set up console spy for dashcam visibility
+      setupConsoleSpy(testdriver, context.task.id);
 
-    // Create the log file on the remote machine
-    const shell = testdriver.os === "windows" ? "pwsh" : "sh";
-    const logPath =
-      testdriver.os === "windows"
-        ? "C:\\Users\\testdriver\\Documents\\testdriver.log"
-        : "/tmp/testdriver.log";
+      // Create the log file on the remote machine
+      const shell = testdriver.os === "windows" ? "pwsh" : "sh";
+      const logPath =
+        testdriver.os === "windows"
+          ? "C:\\Users\\testdriver\\Documents\\testdriver.log"
+          : "/tmp/testdriver.log";
 
-    const createLogCmd =
-      testdriver.os === "windows"
-        ? `New-Item -ItemType File -Path "${logPath}" -Force | Out-Null`
-        : `touch ${logPath}`;
+      const createLogCmd =
+        testdriver.os === "windows"
+          ? `New-Item -ItemType File -Path "${logPath}" -Force | Out-Null`
+          : `touch ${logPath}`;
 
-    await testdriver.exec(shell, createLogCmd, 10000, true);
+      await testdriver.exec(shell, createLogCmd, 10000, true);
 
-    // Only set up dashcam if enabled (default: true)
-    if (testdriver.dashcamEnabled) {
       // Add testdriver log to dashcam tracking
       await testdriver.dashcam.addFileLog(logPath, "TestDriver Log");
 
-      // Start dashcam recording (always, regardless of provision method)
+      // Start dashcam recording
       await testdriver.dashcam.start();
     }
   })();
@@ -335,6 +336,59 @@ export function TestDriver(context, options = {}) {
       return; // Already cleaned up
     }
 
+    // Check if debug-on-failure mode is enabled and test failed
+    const debugOnFailure = currentInstance._debugOnFailure;
+    const testFailed = context.task.result?.state === "fail";
+
+    if (debugOnFailure && testFailed) {
+      // Get sandbox ID before we skip cleanup
+      const instance = currentInstance.getInstance?.();
+      const sandboxId = instance?.sandboxId || instance?.instanceId;
+
+      console.log("");
+      console.log(
+        chalk.yellow(
+          "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
+        ),
+      );
+      console.log(
+        chalk.yellow("  DEBUG MODE: Sandbox kept alive for debugging"),
+      );
+      console.log(
+        chalk.yellow(
+          "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
+        ),
+      );
+      console.log("");
+      console.log("  To connect via MCP:");
+      console.log(
+        chalk.cyan(`    session_start({ sandboxId: "${sandboxId}" })`),
+      );
+      console.log("");
+      console.log(chalk.dim("  Sandbox will expire in 5 minutes."));
+      console.log(
+        chalk.yellow(
+          "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
+        ),
+      );
+      console.log("");
+
+      // Still stop dashcam to save the recording
+      if (currentInstance._dashcam && currentInstance._dashcam.recording) {
+        try {
+          await currentInstance.dashcam.stop();
+        } catch {
+          // Ignore dashcam stop errors in debug mode
+        }
+      }
+
+      // Clean up console spies
+      cleanupConsoleSpy(currentInstance);
+
+      // DO NOT disconnect or terminate - keep sandbox alive for debugging
+      return;
+    }
+
     try {
       // Ensure meta object exists
       if (!context.task.meta) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "testdriverai",
-  "version": "7.2.64",
+  "version": "7.2.65",
   "description": "Next generation autonomous AI agent for end-to-end testing of web & desktop",
   "main": "sdk.js",
   "types": "sdk.d.ts",
@@ -30,16 +30,18 @@
     "./presets": "./lib/presets/index.mjs"
   },
   "bin": {
-    "testdriverai": "bin/testdriverai.js"
+    "testdriverai": "bin/testdriverai.js",
+    "testdriverai-mcp": "mcp-server/dist/server.js"
   },
   "scripts": {
     "start": "node bin/testdriverai.js",
     "dev": "DEV=true node bin/testdriverai.js",
     "debug": "DEV=true VERBOSE=true node bin/testdriverai.js",
-    "docs": "cd docs && npx mint@latest dev",
+    "docs": "npm run docs:skills && cd docs && npx mint@latest dev",
     "docs:dev": "cd docs && npx mint dev",
-    "docs:build": "cd docs && npx mint@latest build",
+    "docs:build": "npm run docs:skills && cd docs && npx mint@latest build",
     "docs:links": "node docs/_scripts/link-replacer.js",
+    "docs:skills": "node docs/_scripts/generate-skills.js",
     "bundle": "node build.mjs",
     "test": "mocha test/*",
     "test:sdk": "vitest run",
@@ -58,11 +60,14 @@
     "test-assert": "node index.js run --path=testdriver/acceptance/assert.yaml",
     "generate-report": "junit-viewer --results=out.xml --save=report.html",
     "serve-report": "npx http-server . -p 8080 -o report.html",
-    "report": "npm run generate-report && npm run serve-report"
+    "report": "npm run generate-report && npm run serve-report",
+    "generate:skills": "node scripts/generate-skills.js"
   },
   "author": "",
   "license": "ISC",
   "dependencies": {
+    "@modelcontextprotocol/ext-apps": "^1.0.0",
+    "@modelcontextprotocol/sdk": "^1.24.0",
     "@npmcli/redact": "^3.2.2",
     "@oclif/core": "^4.5.0",
     "@oclif/plugin-commands": "^4.1.28",
@@ -96,7 +101,8 @@
     "tmp": "^0.2.3",
     "winston": "^3.13.0",
     "winston-transport": "^4.9.0",
-    "ws": "^8.18.1"
+    "ws": "^8.18.1",
+    "zod": "^3.24.0"
   },
   "overrides": {
     "glob": "^11.0.1",
@@ -124,7 +130,8 @@
     "mocha": "^10.8.2",
     "node-addon-api": "^8.0.0",
     "prettier": "3.3.3",
-    "vitest": "^4.0.16"
+    "testdriverai": "^7.2.60",
+    "vitest": "^4.0.18"
   },
   "optionalDependencies": {
     "@esbuild/linux-x64": "^0.21.5"
@@ -137,6 +144,19 @@
       "@oclif/plugin-*"
     ],
     "topicSeparator": " ",
-    "default": "edit"
-  }
+    "default": "init"
+  },
+  "files": [
+    "agent/",
+    "bin/",
+    "ai/skills/",
+    "ai/agents/",
+    "interfaces/",
+    "lib/",
+    "mcp-server/dist/",
+    "sdk.js",
+    "sdk.d.ts",
+    "index.js",
+    "schema.json"
+  ]
 }

package/sdk.d.ts

@@ -14,6 +14,13 @@ export type ScrollDirection = "up" | "down" | "left" | "right";
 export type ScrollMethod = "keyboard" | "mouse";
 export type TextMatchMethod = "ai" | "turbo";
 export type ExecLanguage = "js" | "pwsh" | "sh";
+/**
+ * Preview mode for live test visualization
+ * - "browser": Opens debugger in default browser (default)
+ * - "ide": Opens preview in IDE panel (VSCode, Cursor, etc.)
+ * - "none": Headless mode, no visual preview
+ */
+export type PreviewMode = "browser" | "ide" | "none";
 export type KeyboardKey =
   | "\t"
   | "\n"
@@ -232,7 +239,17 @@ export interface TestDriverOptions {
   };
   /** Force creation of a new sandbox (default: true) */
   newSandbox?: boolean;
-  /** Run in headless mode (default: false) */
+  /**
+   * Preview mode for live test visualization (default: "browser")
+   * - "browser": Opens debugger in default browser
+   * - "ide": Opens preview in IDE panel (VSCode, Cursor, etc.)
+   * - "none": Headless mode, no visual preview
+   */
+  preview?: PreviewMode;
+  /**
+   * @deprecated Use `preview: "none"` instead. Run in headless mode (default: false)
+   * For backward compatibility: headless: true maps to preview: "none"
+   */
   headless?: boolean;
   /** Direct IP address to connect to a running sandbox instance */
   ip?: string;
@@ -280,7 +297,17 @@ export interface ConnectOptions {
   sandboxInstance?: string;
   /** Operating system for the sandbox (default: 'linux') */
   os?: "windows" | "linux";
-  /** Run in headless mode (default: false) */
+  /**
+   * Preview mode for live test visualization (default: "browser")
+   * - "browser": Opens debugger in default browser
+   * - "ide": Opens preview in IDE panel (VSCode, Cursor, etc.)
+   * - "none": Headless mode, no visual preview
+   */
+  preview?: PreviewMode;
+  /**
+   * @deprecated Use `preview: "none"` instead. Run in headless mode (default: false)
+   * For backward compatibility: headless: true maps to preview: "none"
+   */
   headless?: boolean;
   /** Reuse recent connection if available (default: true) */
   reuseConnection?: boolean;

package/sdk.js

@@ -3,6 +3,7 @@ const path = require("path");
 const os = require("os");
 const crypto = require("crypto");
 const { formatter } = require("./sdk-log-formatter");
+const logger = require("./agent/lib/logger");
 
 /**
  * Get the file path of the caller (the file that called TestDriver)

package/.env.example

@@ -1,4 +0,0 @@
-TD_API_KEY=
-TD_API_ROOT=
-
-TWOCAPTCHA_API_KEY=

package/.github/workflows/acceptance-linux-scheduled.yaml

@@ -1,45 +0,0 @@
-name: Scheduled Acceptance Tests (Linux)
-
-permissions:
-  id-token: write
-  contents: read
-
-on:
-  schedule:
-    # Every 6 hours
-    - cron: "0 */6 * * *"
-  workflow_dispatch: # Allow manual trigger
-
-jobs:
-  test-linux:
-    runs-on: ubuntu-latest
-
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Setup Node.js
-        uses: actions/setup-node@v4
-        with:
-          node-version: "20"
-          cache: "npm"
-
-      - name: Install dependencies
-        run: npm ci
-
-      - name: Install Sentry CLI
-        run: npm install -g @sentry/cli
-
-      - name: Run Linux tests with Sentry Cron monitoring
-        run: |
-          sentry-cli monitors run testdriver-linux-acceptance -- npx vitest run examples/*.test.mjs
-        env:
-          SENTRY_DSN: ${{ secrets.SENTRY_DSN }}
-          TD_API_KEY: ${{ secrets.TD_API_KEY }}
-          TD_OS: linux
-          TWOCAPTCHA_API_KEY: ${{ secrets.TWOCAPTCHA_API_KEY }}
-
-      - name: Upload test results to Sentry Prevent
-        if: ${{ !cancelled() }}
-        uses: getsentry/prevent-action@v0
-        with:
-          token: ${{ secrets.SENTRY_PREVENT_TOKEN }}

package/.github/workflows/acceptance-windows-scheduled.yaml

@@ -1,54 +0,0 @@
-name: Scheduled Acceptance Tests (Windows)
-
-permissions:
-  id-token: write
-  contents: read
-
-on:
-  schedule:
-    # Every day at midnight UTC
-    - cron: "0 0 * * *"
-  workflow_dispatch: # Allow manual trigger
-
-jobs:
-  test-windows:
-    uses: ./.github/workflows/windows-self-hosted.yaml
-    secrets:
-      TD_API_KEY: ${{ secrets.TD_API_KEY }}
-      TWOCAPTCHA_API_KEY: ${{ secrets.TWOCAPTCHA_API_KEY }}
-      TD_WEBSITE: ${{ secrets.TD_WEBSITE }}
-      AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
-      AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
-
-  sentry-and-publish:
-    needs: test-windows
-    runs-on: ubuntu-latest
-    if: always()
-
-    steps:
-      - name: Install Sentry CLI
-        run: npm install -g @sentry/cli
-
-      - name: Report to Sentry Cron
-        run: |
-          sentry-cli monitors run testdriver-windows-acceptance -- echo "Windows tests completed"
-        env:
-          SENTRY_DSN: ${{ secrets.SENTRY_DSN }}
-
-      - name: Download test results
-        uses: actions/download-artifact@v4
-        with:
-          name: test-results-windows
-
-      - name: Upload test results to Sentry Prevent
-        if: ${{ !cancelled() }}
-        uses: getsentry/prevent-action@v0
-        with:
-          token: ${{ secrets.SENTRY_PREVENT_TOKEN }}
-
-      - name: Publish Test Results
-        uses: EnricoMi/publish-unit-test-result-action@v2
-        if: always()
-        with:
-          files: test-report.junit.xml
-          check_name: Test Results (Windows Scheduled)

package/.github/workflows/acceptance.yaml

@@ -1,87 +0,0 @@
-name: Acceptance Tests
-permissions:
-  id-token: write
-  contents: write
-  pull-requests: write
-  checks: write
-on:
-  pull_request:
-    branches: [main]
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
-  cancel-in-progress: true
-
-jobs:
-  test-linux:
-    runs-on: ubuntu-latest
-
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Setup Node.js
-        uses: actions/setup-node@v4
-        with:
-          node-version: "20"
-          cache: "npm"
-
-      - name: Install dependencies
-        run: npm ci
-
-      - name: Run Linux tests
-        run: npx vitest run examples/*.test.mjs
-        env:
-          TD_API_KEY: ${{ secrets.TD_API_KEY }}
-          TWOCAPTCHA_API_KEY: ${{ secrets.TWOCAPTCHA_API_KEY }}
-          TD_OS: linux
-
-      - name: Upload test results to Sentry Prevent
-        if: ${{ !cancelled() }}
-        uses: getsentry/prevent-action@v0
-        with:
-          token: ${{ secrets.SENTRY_PREVENT_TOKEN }}
-
-      - name: Publish Test Results
-        uses: EnricoMi/publish-unit-test-result-action@v2
-        if: always()
-        with:
-          files: test-report.junit.xml
-          comment_mode: always
-          check_name: Test Results (Linux)
-
-  test-windows:
-    if: contains(github.event.pull_request.labels.*.name, 'test-windows')
-    uses: ./.github/workflows/windows-self-hosted.yaml
-    with:
-      test_pattern: "examples/*.test.mjs"
-    secrets:
-      TD_API_KEY: ${{ secrets.TD_API_KEY }}
-      TWOCAPTCHA_API_KEY: ${{ secrets.TWOCAPTCHA_API_KEY }}
-      TD_WEBSITE: ${{ secrets.TD_WEBSITE }}
-      AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
-      AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
-
-  publish-windows-results:
-    needs: test-windows
-    runs-on: ubuntu-latest
-    if: always() && contains(github.event.pull_request.labels.*.name, 'test-windows')
-
-    steps:
-      - name: Download test results
-        uses: actions/download-artifact@v4
-        with:
-          name: test-results-windows
-
-      - name: Upload test results to Sentry Prevent
-        if: ${{ !cancelled() }}
-        uses: getsentry/prevent-action@v0
-        with:
-          token: ${{ secrets.SENTRY_PREVENT_TOKEN }}
-
-      - name: Publish Test Results
-        uses: EnricoMi/publish-unit-test-result-action@v2
-        if: always()
-        with:
-          files: test-report.junit.xml
-          comment_mode: always
-          check_name: Test Results (Windows)

package/.github/workflows/publish.yaml

@@ -1,68 +0,0 @@
-name: Publish
-permissions:
-  contents: write
-  id-token: write # Required for OIDC
-on:
-  push:
-    branches: [main]
-
-jobs:
-  publish:
-    runs-on: ubuntu-latest
-
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          token: ${{ secrets.GITHUB_TOKEN }}
-
-      - name: Setup Node.js
-        uses: actions/setup-node@v4
-        with:
-          node-version: "20"
-          registry-url: "https://registry.npmjs.org/"
-
-      - name: Configure Git
-        run: |
-          git config user.name "github-actions[bot]"
-          git config user.email "github-actions[bot]@users.noreply.github.com"
-
-      - name: Install dependencies
-        run: npm ci
-
-      - name: Bump version (patch)
-        run: npm version patch --no-git-tag-version
-
-      - name: Generate Changelog
-        run: |
-          npx conventional-changelog-cli -p angular -i CHANGELOG.md -s
-          git add CHANGELOG.md
-
-      - name: Commit and push version bump
-        run: |
-          git add package.json package-lock.json
-          VERSION=$(node -p "require('./package.json').version")
-          git commit -m "chore(release): ${VERSION}"
-          git tag "v${VERSION}"
-          git push origin main
-          git push origin "v${VERSION}"
-          echo "VERSION=${VERSION}" >> $GITHUB_ENV
-
-      - name: Create GitHub Release
-        uses: softprops/action-gh-release@v1
-        with:
-          tag_name: v${{ env.VERSION }}
-          generate_release_notes: true
-          prerelease: false
-
-      - name: Debug NPM Token
-        run: |
-          echo "NPM_TOKEN is set: ${{ secrets.NPM_TOKEN != '' }}"
-          echo "NPM_TOKEN first 4 chars: ${NPM_TOKEN:0:4}..."
-        env:
-          NPM_TOKEN: ${{ secrets.NPM_TOKEN }}
-
-      - name: Publish to npm
-        run: npm publish --tag beta
-        env:
-          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}