testdriverai 7.8.0-test.7 → 7.8.0-test.71

package/{examples → manual}/drag-and-drop.test.mjs

@@ -46,7 +46,7 @@ describe("Drag and Drop Test", () => {
 
       const recycleBin = await testdriver.find(
         "Recycle Bin, recycle bin icon in the top left corner of the desktop",
-      );
+      ).hover();
       await recycleBin.mouseUp();
 
       // Assert "New Text Document" icon is not on the Desktop

package/manual/exec-stream-logs.test.mjs

@@ -0,0 +1,25 @@
+import { describe, expect, it } from "vitest";
+import { TestDriver } from "../lib/vitest/hooks.mjs";
+import { getDefaults } from "../examples/config.mjs";
+
+describe("Exec Log Streaming", () => {
+  it("should stream exec logs every second for 20 seconds", async (context) => {
+    const testdriver = TestDriver(context, { ...getDefaults(context), headless: true });
+    await testdriver.provision.chrome({ url: "about:blank" });
+
+    const code = `for i in $(seq 1 20); do echo "log line $i at $(date +%T)"; sleep 1; done`;
+
+    const result = await testdriver.exec({
+      language: "sh",
+      code,
+      timeout: 30000,
+    });
+
+    console.log("exec result:", result);
+
+    // Verify we got all 20 log lines
+    for (let i = 1; i <= 20; i++) {
+      expect(result).toContain(`log line ${i}`);
+    }
+  });
+});

package/mcp-server/dist/server.mjs

@@ -26,18 +26,33 @@ import { sessionManager } from "./session.js";
 const sdkRoot = path.join(path.dirname(fileURLToPath(import.meta.url)), "..", "..");
 const packageJson = JSON.parse(fs.readFileSync(path.join(sdkRoot, "package.json"), "utf-8"));
 const version = packageJson.version || "1.0.0";
-// Derive release channel from package version prerelease tag (e.g. "7.6.0-test.5" → "test")
+// Derive release channel and infrastructure environment from package version
 import semver from "semver";
-const KNOWN_CHANNELS = new Set(["dev", "test", "canary", "latest"]);
-function resolveReleaseChannel(ver) {
-    if (process.env.TD_CHANNEL && KNOWN_CHANNELS.has(process.env.TD_CHANNEL))
+const CHANNEL_TO_ENV = {
+    dev: "dev",
+    test: "staging",
+    canary: "production",
+    stable: "production",
+};
+const VALID_CHANNELS = new Set(Object.keys(CHANNEL_TO_ENV));
+const VALID_ENVS = new Set(["dev", "staging", "production"]);
+function resolveChannel(ver) {
+    if (process.env.TD_CHANNEL && VALID_CHANNELS.has(process.env.TD_CHANNEL))
         return process.env.TD_CHANNEL;
+    if (process.env.TD_ENV && VALID_CHANNELS.has(process.env.TD_ENV))
+        return process.env.TD_ENV;
     const pre = semver.prerelease(ver);
-    if (pre && pre.length > 0 && KNOWN_CHANNELS.has(String(pre[0])))
+    if (pre && pre.length > 0 && VALID_CHANNELS.has(String(pre[0])))
         return String(pre[0]);
-    return "latest";
+    return "stable";
+}
+function resolveSentryEnvironment(ver) {
+    if (process.env.TD_ENV && VALID_ENVS.has(process.env.TD_ENV))
+        return process.env.TD_ENV;
+    return CHANNEL_TO_ENV[resolveChannel(ver)] || "production";
 }
-const releaseChannel = resolveReleaseChannel(version);
+const activeChannel = resolveChannel(version);
+const sentryEnvironment = resolveSentryEnvironment(version);
 const isSentryEnabled = () => {
     if (process.env.TD_TELEMETRY === "false") {
         return false;
@@ -49,7 +64,7 @@ if (isSentryEnabled()) {
     Sentry.init({
         dsn: process.env.SENTRY_DSN ||
             "https://452bd5a00dbd83a38ee8813e11c57694@o4510262629236736.ingest.us.sentry.io/4510480443637760",
-        environment: releaseChannel,
+        environment: sentryEnvironment,
         release: version,
         sampleRate: 1.0,
         tracesSampleRate: 1.0,
@@ -57,6 +72,7 @@ if (isSentryEnabled()) {
         integrations: [Sentry.httpIntegration(), Sentry.nodeContextIntegration()],
         initialScope: {
             tags: {
+                channel: activeChannel,
                 platform: os.platform(),
                 arch: os.arch(),
                 nodeVersion: process.version,
@@ -84,6 +100,10 @@ if (isSentryEnabled()) {
             if (error && typeof error === "object" && "name" in error && error.name === "TestFailure") {
                 return null;
             }
+            // Filter out ElementNotFoundError - expected test outcome, not a crash
+            if (error && typeof error === "object" && "name" in error && error.name === "ElementNotFoundError") {
+                return null;
+            }
             return event;
         },
     });

package/mcp-server/src/server.ts

@@ -34,16 +34,33 @@ const sdkRoot = path.join(path.dirname(fileURLToPath(import.meta.url)), "..", ".
 const packageJson = JSON.parse(fs.readFileSync(path.join(sdkRoot, "package.json"), "utf-8"));
 const version = packageJson.version || "1.0.0";
 
-// Derive release channel from package version prerelease tag (e.g. "7.6.0-test.5" → "test")
+// Derive release channel and infrastructure environment from package version
 import semver from "semver";
-const KNOWN_CHANNELS = new Set(["dev", "test", "canary", "latest"]);
-function resolveReleaseChannel(ver: string): string {
-  if (process.env.TD_CHANNEL && KNOWN_CHANNELS.has(process.env.TD_CHANNEL)) return process.env.TD_CHANNEL;
+
+const CHANNEL_TO_ENV: Record<string, string> = {
+  dev: "dev",
+  test: "staging",
+  canary: "production",
+  stable: "production",
+};
+const VALID_CHANNELS = new Set(Object.keys(CHANNEL_TO_ENV));
+const VALID_ENVS = new Set(["dev", "staging", "production"]);
+
+function resolveChannel(ver: string): string {
+  if (process.env.TD_CHANNEL && VALID_CHANNELS.has(process.env.TD_CHANNEL)) return process.env.TD_CHANNEL;
+  if (process.env.TD_ENV && VALID_CHANNELS.has(process.env.TD_ENV)) return process.env.TD_ENV;
   const pre = semver.prerelease(ver);
-  if (pre && pre.length > 0 && KNOWN_CHANNELS.has(String(pre[0]))) return String(pre[0]);
-  return "latest";
+  if (pre && pre.length > 0 && VALID_CHANNELS.has(String(pre[0]))) return String(pre[0]);
+  return "stable";
+}
+
+function resolveSentryEnvironment(ver: string): string {
+  if (process.env.TD_ENV && VALID_ENVS.has(process.env.TD_ENV)) return process.env.TD_ENV;
+  return CHANNEL_TO_ENV[resolveChannel(ver)] || "production";
 }
-const releaseChannel = resolveReleaseChannel(version);
+
+const activeChannel = resolveChannel(version);
+const sentryEnvironment = resolveSentryEnvironment(version);
 
 const isSentryEnabled = () => {
   if (process.env.TD_TELEMETRY === "false") {
@@ -58,7 +75,7 @@ if (isSentryEnabled()) {
     dsn:
       process.env.SENTRY_DSN ||
       "https://452bd5a00dbd83a38ee8813e11c57694@o4510262629236736.ingest.us.sentry.io/4510480443637760",
-    environment: releaseChannel,
+    environment: sentryEnvironment,
     release: version,
     sampleRate: 1.0,
     tracesSampleRate: 1.0,
@@ -66,6 +83,7 @@ if (isSentryEnabled()) {
     integrations: [Sentry.httpIntegration(), Sentry.nodeContextIntegration()],
     initialScope: {
       tags: {
+        channel: activeChannel,
         platform: os.platform(),
         arch: os.arch(),
         nodeVersion: process.version,
@@ -99,6 +117,11 @@ if (isSentryEnabled()) {
       if (error && typeof error === "object" && "name" in error && (error as { name: string }).name === "TestFailure") {
         return null;
       }
+
+      // Filter out ElementNotFoundError - expected test outcome, not a crash
+      if (error && typeof error === "object" && "name" in error && (error as { name: string }).name === "ElementNotFoundError") {
+        return null;
+      }
       
       return event;
     },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "testdriverai",
-  "version": "7.8.0-test.7",
+  "version": "7.8.0-test.71",
   "description": "Next generation autonomous AI agent for end-to-end testing of web & desktop",
   "main": "sdk.js",
   "types": "sdk.d.ts",
@@ -37,8 +37,8 @@
     "start": "node bin/testdriverai.js",
     "dev": "DEV=true node bin/testdriverai.js",
     "debug": "DEV=true VERBOSE=true node bin/testdriverai.js",
-    "docs": "npm run docs:skills && cd docs && npx mint@latest dev",
-    "docs:dev": "cd docs && npx mint dev",
+    "docs": "npm run docs:skills && cd docs && npx mint@latest dev --port 3002",
+    "docs:dev": "cd docs && npx mint dev --port 3002",
     "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",
@@ -116,6 +116,7 @@
   },
   "overrides": {
     "glob": "^11.0.1",
+    "obug": "2.1.0",
     "rimraf": "^5.0.10"
   },
   "peerDependencies": {

package/sdk.d.ts

@@ -273,6 +273,8 @@ export interface TestDriverOptions {
   sandboxAmi?: string;
   /** EC2 instance type for sandbox (e.g., 'i3.metal') */
   sandboxInstance?: string;
+  /** E2B template ID to use when creating the sandbox (e.g., 'my-template-id') */
+  e2bTemplateId?: string;
   /** Cache key for element finding operations. If provided, enables caching tied to this key */
   cacheKey?: string;
   /** Reconnect to the last used sandbox instead of creating a new one. When true, provision methods (chrome, vscode, installer, etc.) will be skipped since the application is already running. Throws error if no previous sandbox exists. */
@@ -327,6 +329,8 @@ export interface ConnectOptions {
   sandboxAmi?: string;
   /** EC2 instance type for sandbox (e.g., 'i3.metal') */
   sandboxInstance?: string;
+  /** E2B template ID to use when creating the sandbox (e.g., 'my-template-id') */
+  e2bTemplateId?: string;
   /** Operating system for the sandbox (default: 'linux') */
   os?: "windows" | "linux";
   /**

package/sdk.js

@@ -481,7 +481,7 @@ class Element {
       let cacheKey = null;
       let cacheThreshold = null;
       let perCommandThresholds = null; // Per-command { screen, element } override
-      let zoom = false; // Default to disabled, enable with zoom: true
+      let zoom = true; // Default to enabled
       let perCommandAi = null; // Per-command AI config override
 
       let minConfidence = null; // Minimum confidence threshold
@@ -494,8 +494,8 @@ class Element {
         // New: options is an object with cacheKey and/or cacheThreshold
         cacheKey = options.cacheKey || null;
         cacheThreshold = options.cacheThreshold ?? null;
-        // zoom defaults to false unless explicitly set to true
-        zoom = options.zoom === true;
+        // zoom defaults to true unless explicitly set to false
+        zoom = options.zoom !== false;
         // Minimum confidence threshold: fail find if AI confidence is below this value
         minConfidence = options.confidence ?? null;
         // Element type hint for prompt wrapping
@@ -568,7 +568,7 @@ class Element {
         cacheKey: cacheKey,
         os: this.sdk.os,
         resolution: this.sdk.resolution,
-        zoom: zoom,
+        zoom: zoom === true ? 1 : zoom === false ? 0 : zoom,
         confidence: minConfidence,
         type: elementType,
         ai: {
@@ -623,6 +623,11 @@ class Element {
 
     // Track find interaction once at the end (fire-and-forget, don't block)
     const sessionId = this.sdk.getSessionId();
+    const findCacheHit = response?.cacheHit || response?.cache_hit || response?.cached || false;
+    // Increment local interaction counters
+    this.sdk._interactionStats.total++;
+    this.sdk._interactionStats.byType.find = (this.sdk._interactionStats.byType.find || 0) + 1;
+    if (findCacheHit) this.sdk._interactionStats.cached++;
     if (sessionId && this.sdk.apiClient) {
       this.sdk.apiClient
         .req("interaction/track", {
@@ -632,11 +637,7 @@ class Element {
           timestamp: absoluteTimestamp, // Absolute epoch timestamp - frontend calculates relative using clientStartDate
           success: this._found,
           error: findError,
-          cacheHit:
-            response?.cacheHit ||
-            response?.cache_hit ||
-            response?.cached ||
-            false,
+          cacheHit: findCacheHit,
           selector: response?.selector,
           selectorUsed: !!response?.selector,
           confidence: response?.confidence ?? null,
@@ -1498,6 +1499,7 @@ class TestDriverSDK {
     // Store sandbox configuration options
     this.sandboxAmi = options.sandboxAmi || null;
     this.sandboxInstance = options.sandboxInstance || null;
+    this.e2bTemplateId = options.e2bTemplateId || null;
 
     // Store reconnect preference from options
     this.reconnect =
@@ -1616,6 +1618,12 @@ class TestDriverSDK {
     // Uploaded to S3 at cleanup so they can be displayed alongside dashcam replays.
     this._logBuffer = [];
 
+    // API version discovered by _logEnvironmentInfo()
+    this._apiVersion = null;
+
+    // Local interaction counters — incremented at each interaction/track call site
+    this._interactionStats = { total: 0, cached: 0, byType: {} };
+
     // Set up event listeners once (they live for the lifetime of the SDK instance)
     this._setupLogging();
 
@@ -2716,6 +2724,7 @@ CAPTCHA_SOLVER_EOF`,
    * @param {string} options.ip - Direct IP address to connect to
    * @param {string} options.sandboxAmi - AMI to use for the sandbox
    * @param {string} options.sandboxInstance - Instance type for the sandbox
+   * @param {string} options.e2bTemplateId - E2B template ID to use when creating the sandbox
    * @param {string} options.os - Operating system for the sandbox (windows or linux)
    * @param {boolean} options.reuseConnection - Reuse recent connection if available (default: true)
    * @returns {Promise<Object>} Sandbox instance details
@@ -2749,6 +2758,9 @@ CAPTCHA_SOLVER_EOF`,
       }
     }
 
+    // Log environment info immediately so it's visible even if auth fails
+    this._logEnvironmentInfo();
+
     // Authenticate first if not already authenticated
     if (!this.authenticated) {
       await this.auth();
@@ -2801,6 +2813,12 @@ CAPTCHA_SOLVER_EOF`,
     } else if (this.sandboxInstance) {
       this.agent.sandboxInstance = this.sandboxInstance;
     }
+    // Use e2bTemplateId from connectOptions if provided, otherwise fall back to constructor value
+    if (connectOptions.e2bTemplateId !== undefined) {
+      this.agent.e2bTemplateId = connectOptions.e2bTemplateId;
+    } else if (this.e2bTemplateId) {
+      this.agent.e2bTemplateId = this.e2bTemplateId;
+    }
     // Use os from connectOptions if provided, otherwise fall back to this.os
     if (connectOptions.os !== undefined) {
       this.agent.sandboxOs = connectOptions.os;
@@ -2865,9 +2883,6 @@ CAPTCHA_SOLVER_EOF`,
       sandboxId: this.instance?.instanceId,
     });
 
-    // Log environment info (non-blocking, skip on stable)
-    this._logEnvironmentInfo();
-
     return this.instance;
   }
 
@@ -3193,6 +3208,11 @@ CAPTCHA_SOLVER_EOF`,
 
         // Track successful findAll interaction (fire-and-forget, don't block)
         const sessionId = this.getSessionId();
+        const findAllCacheHit = response.cached || false;
+        // Increment local interaction counters
+        this._interactionStats.total++;
+        this._interactionStats.byType.findAll = (this._interactionStats.byType.findAll || 0) + 1;
+        if (findAllCacheHit) this._interactionStats.cached++;
         if (sessionId && this.apiClient) {
           this.apiClient
             .req("interaction/track", {
@@ -3202,7 +3222,7 @@ CAPTCHA_SOLVER_EOF`,
               timestamp: absoluteTimestamp, // Absolute epoch timestamp - frontend calculates relative using clientStartDate
               success: true,
               input: { count: elements.length },
-              cacheHit: response.cached || false,
+              cacheHit: findAllCacheHit,
               selector: response.selector,
               selectorUsed: !!response.selector,
               screenshotUrl: response.screenshotKey ?? null,
@@ -3248,6 +3268,11 @@ CAPTCHA_SOLVER_EOF`,
 
         // No elements found - track interaction (fire-and-forget, don't block)
         const sessionId = this.getSessionId();
+        const noResultCacheHit = response?.cached || false;
+        // Increment local interaction counters
+        this._interactionStats.total++;
+        this._interactionStats.byType.findAll = (this._interactionStats.byType.findAll || 0) + 1;
+        if (noResultCacheHit) this._interactionStats.cached++;
         if (sessionId && this.apiClient) {
           this.apiClient
             .req("interaction/track", {
@@ -3258,7 +3283,7 @@ CAPTCHA_SOLVER_EOF`,
               success: false,
               error: "No elements found",
               input: { count: 0 },
-              cacheHit: response?.cached || false,
+              cacheHit: noResultCacheHit,
               selector: response?.selector,
               selectorUsed: !!response?.selector,
               screenshotUrl: response?.screenshotKey ?? null,
@@ -3292,6 +3317,9 @@ CAPTCHA_SOLVER_EOF`,
 
       // Track findAll error interaction (fire-and-forget, don't block)
       const sessionId = this.getSessionId();
+      // Increment local interaction counters
+      this._interactionStats.total++;
+      this._interactionStats.byType.findAll = (this._interactionStats.byType.findAll || 0) + 1;
       if (sessionId && this.apiClient) {
         this.apiClient
           .req("interaction/track", {
@@ -3817,7 +3845,7 @@ CAPTCHA_SOLVER_EOF`,
     const apiRoot = this.config?.TD_API_ROOT || 'unknown';
     const apiKey = this.config?.TD_API_KEY || '';
     const maskedKey = apiKey.length > 4 ? '***' + apiKey.slice(-4) : '(not set)';
-    const env = process.env.TD_ENV || 'unknown';
+    const env = process.env.TD_CHANNEL || process.env.TD_ENV || 'unknown';
     const os = this.agent?.options?.os || process.env.TD_OS || 'linux';
     const sdkVersion = require('./package.json').version;
 
@@ -3843,6 +3871,8 @@ CAPTCHA_SOLVER_EOF`,
       res.on('end', () => {
         try {
           const info = JSON.parse(data);
+          // Persist API version for test result metadata
+          this._apiVersion = info.version || null;
           const commit = info.commit || 'unknown';
           const shortCommit = commit.substring(0, 7);
           const commitUrl = commit !== 'unknown'

package/setup/aws/install-dev-runner.sh

@@ -0,0 +1,79 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# Usage: ./install-dev-runner.sh <instance-id>
+INSTANCE_ID="${1:?Usage: $0 <instance-id>}"
+AWS_REGION="${AWS_REGION:-us-east-2}"
+
+RUNNER_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../../../runner" && pwd)"
+
+echo "Packing local runner..."
+TMPDIR=$(mktemp -d)
+pushd "$RUNNER_DIR" > /dev/null
+npm pack --pack-destination "$TMPDIR" > /dev/null 2>&1
+TARBALL=$(ls "$TMPDIR"/*.tgz)
+popd > /dev/null
+echo "Tarball: $TARBALL"
+
+echo "Uploading to S3..."
+S3_KEY="runner-dev/$(date +%s)-$(openssl rand -hex 4)/runner.tgz"
+aws s3 cp "$TARBALL" "s3://v7-transfer/${S3_KEY}" --region "$AWS_REGION" > /dev/null
+DOWNLOAD_URL=$(aws s3 presign "s3://v7-transfer/${S3_KEY}" --expires-in 900 --region "$AWS_REGION")
+rm -rf "$TMPDIR"
+
+echo "Creating SSM params file..."
+
+# Write Python script to temp file to generate valid JSON
+PYTHON_SCRIPT=$(mktemp --suffix=.py)
+cat > "$PYTHON_SCRIPT" << 'PYEOF'
+import json
+import sys
+
+url = sys.argv[1]
+
+commands = [
+    "Write-Host '=== Stopping runner ==='",
+    "Stop-ScheduledTask -TaskName RunTestDriverAgent -ErrorAction SilentlyContinue",
+    "Stop-Process -Name node -Force -ErrorAction SilentlyContinue", 
+    "Start-Sleep -Seconds 2",
+    "Set-Location 'C:\\testdriver\\sandbox-agent'",
+    "$tarball = 'C:\\Windows\\Temp\\runner-dev.tgz'",
+    f"Invoke-WebRequest -Uri '{url}' -OutFile $tarball",
+    "Write-Host 'Tarball size:'; (Get-Item $tarball).Length",
+    "Remove-Item -Path lib -Recurse -Force -ErrorAction SilentlyContinue", 
+    "tar -xzf $tarball --strip-components=1 -C .",
+    "Get-Content 'package.json' | ConvertFrom-Json | Select-Object -ExpandProperty version",
+    "Write-Host '=== Starting runner ==='",
+    "Start-ScheduledTask -TaskName RunTestDriverAgent",
+    "Start-Sleep -Seconds 3",
+    "Get-Content 'C:\\testdriver\\log.txt' -Tail 20"
+]
+
+params = {"commands": commands}
+print(json.dumps(params))
+PYEOF
+
+python3 "$PYTHON_SCRIPT" "$DOWNLOAD_URL" > /tmp/ssm-install-params.json
+rm "$PYTHON_SCRIPT"
+
+echo "Sending SSM command..."
+CMD_JSON=$(aws ssm send-command \
+  --region "$AWS_REGION" \
+  --instance-ids "$INSTANCE_ID" \
+  --document-name "AWS-RunPowerShellScript" \
+  --parameters "file:///tmp/ssm-install-params.json" \
+  --output json)
+
+COMMAND_ID=$(echo "$CMD_JSON" | jq -r '.Command.CommandId')
+echo "Command ID: $COMMAND_ID"
+
+echo "Waiting for completion..."
+aws ssm wait command-executed --region "$AWS_REGION" --command-id "$COMMAND_ID" --instance-id "$INSTANCE_ID" || true
+
+echo "Getting output..."
+aws ssm get-command-invocation \
+  --region "$AWS_REGION" \
+  --command-id "$COMMAND_ID" \
+  --instance-id "$INSTANCE_ID" \
+  --query 'StandardOutputContent' \
+  --output text

package/setup/aws/spawn-runner.sh

@@ -141,6 +141,171 @@ while :; do
   sleep 20
 done
 
+# --- 4) Install/update runner ---
+echo "Installing runner..."
+
+# Determine environment and version
+TD_CHANNEL="${TD_CHANNEL:-stable}"
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+SDK_PKG_JSON="${SCRIPT_DIR}/../../../sdk/package.json"
+RUNNER_DIR="${SCRIPT_DIR}/../../../runner"
+
+if [ -f "$SDK_PKG_JSON" ]; then
+  RUNNER_VERSION=$(jq -r '.version' "$SDK_PKG_JSON")
+  echo "Runner version from SDK: $RUNNER_VERSION"
+else
+  RUNNER_VERSION="$TD_CHANNEL"
+  echo "SDK package.json not found, using env tag: $RUNNER_VERSION"
+fi
+
+if [ "$TD_CHANNEL" = "dev" ]; then
+  echo "Dev mode: packing and uploading local runner to S3..."
+  
+  # Pack local runner
+  TMPDIR=$(mktemp -d)
+  pushd "$RUNNER_DIR" > /dev/null
+  npm pack --pack-destination "$TMPDIR" > /dev/null 2>&1
+  TARBALL=$(ls "$TMPDIR"/*.tgz | head -1)
+  popd > /dev/null
+  
+  # Upload to S3
+  S3_BUCKET="${AWS_BUCKET_IMAGE_TRANSFER:-v7-transfer}"
+  S3_KEY="runner-dev/$(date +%s)-$(openssl rand -hex 4)/runner.tgz"
+  aws s3 cp "$TARBALL" "s3://${S3_BUCKET}/${S3_KEY}" --region "$AWS_REGION"
+  
+  # Generate presigned URL (15 min)
+  DOWNLOAD_URL=$(aws s3 presign "s3://${S3_BUCKET}/${S3_KEY}" --expires-in 900 --region "$AWS_REGION")
+  rm -rf "$TMPDIR"
+  
+  # Build SSM parameters JSON in a temp file to avoid shell escaping issues with URL
+  PARAMS_FILE=$(mktemp)
+  cat > "$PARAMS_FILE" << 'PARAMS_EOF'
+{
+  "commands": [
+    "Write-Host '=== Starting runner dev install ==='",
+    "Write-Host 'Stopping existing runner processes...'",
+    "Stop-ScheduledTask -TaskName RunTestDriverAgent -ErrorAction SilentlyContinue",
+    "Stop-Process -Name node -Force -ErrorAction SilentlyContinue",
+    "Start-Sleep -Seconds 2",
+    "Write-Host 'Current runner version:'",
+    "Get-Content 'C:\\testdriver\\sandbox-agent\\package.json' | ConvertFrom-Json | Select-Object -ExpandProperty version",
+    "Set-Location 'C:\\testdriver\\sandbox-agent'",
+    "Write-Host 'Dev mode: downloading runner from S3...'",
+    "$tarball = 'C:\\Windows\\Temp\\runner-dev.tgz'",
+PARAMS_EOF
+  
+  # Add the URL line with proper JSON escaping
+  echo "    \"Invoke-WebRequest -Uri '$(echo "$DOWNLOAD_URL" | sed 's/"/\\"/g')' -OutFile \$tarball\"," >> "$PARAMS_FILE"
+  
+  cat >> "$PARAMS_FILE" << 'PARAMS_EOF'
+    "Write-Host 'Downloaded tarball size:'",
+    "(Get-Item $tarball).Length",
+    "Write-Host 'Extracting runner...'",
+    "tar -xzf $tarball -C 'C:\\Windows\\Temp'",
+    "Write-Host 'Extracted package contents:'",
+    "Get-ChildItem 'C:\\Windows\\Temp\\package' -Recurse | Select-Object FullName",
+    "Write-Host 'New runner version in package:'",
+    "Get-Content 'C:\\Windows\\Temp\\package\\package.json' | ConvertFrom-Json | Select-Object -ExpandProperty version",
+    "Write-Host 'Clearing old lib folder...'",
+    "Remove-Item 'C:\\testdriver\\sandbox-agent\\lib' -Recurse -Force -ErrorAction SilentlyContinue",
+    "Write-Host 'Copying files to sandbox-agent...'",
+    "xcopy 'C:\\Windows\\Temp\\package\\*' 'C:\\testdriver\\sandbox-agent\\' /E /Y /I",
+    "Write-Host 'Files after copy:'",
+    "Get-ChildItem 'C:\\testdriver\\sandbox-agent' | Select-Object Name",
+    "Remove-Item 'C:\\Windows\\Temp\\package' -Recurse -Force -ErrorAction SilentlyContinue",
+    "Remove-Item $tarball -Force -ErrorAction SilentlyContinue",
+    "Write-Host 'Runner version after copy:'",
+    "Get-Content 'C:\\testdriver\\sandbox-agent\\package.json' | ConvertFrom-Json | Select-Object -ExpandProperty version",
+    "Write-Host 'Installing npm dependencies...'",
+    "npm install --omit=dev 2>&1 | Write-Host",
+    "Write-Host 'Final verification - ably-service.js exists:'",
+    "Test-Path 'C:\\testdriver\\sandbox-agent\\lib\\ably-service.js'",
+    "Write-Host 'Restarting RunTestDriverAgent scheduled task...'",
+    "Start-ScheduledTask -TaskName RunTestDriverAgent -ErrorAction SilentlyContinue",
+    "Write-Host '=== Runner install complete (dev) ==='"
+  ]
+}
+PARAMS_EOF
+
+  echo "Sending SSM command to download and install runner from S3..."
+  INSTALL_CMD=$(aws ssm send-command \
+    --region "$AWS_REGION" \
+    --instance-ids "$INSTANCE_ID" \
+    --document-name "AWS-RunPowerShellScript" \
+    --parameters "file://$PARAMS_FILE" \
+    --timeout-seconds 180 \
+    --output json)
+  rm -f "$PARAMS_FILE"
+else
+  echo "Installing @testdriverai/runner@${RUNNER_VERSION} via npm pack + extract..."
+
+  # Build SSM parameters JSON in a temp file (same approach as dev mode)
+  PARAMS_FILE=$(mktemp)
+  cat > "$PARAMS_FILE" << PARAMS_EOF
+{
+  "commands": [
+    "Write-Host '=== Starting runner install (npm pack) ==='",
+    "Write-Host 'Stopping existing runner processes...'",
+    "Stop-ScheduledTask -TaskName RunTestDriverAgent -ErrorAction SilentlyContinue",
+    "Stop-Process -Name node -Force -ErrorAction SilentlyContinue",
+    "Start-Sleep -Seconds 2",
+    "Write-Host 'Current runner version:'",
+    "Get-Content 'C:\\\\testdriver\\\\sandbox-agent\\\\package.json' | ConvertFrom-Json | Select-Object -ExpandProperty version",
+    "Set-Location 'C:\\\\Windows\\\\Temp'",
+    "Write-Host 'Downloading @testdriverai/runner@${RUNNER_VERSION} via npm pack...'",
+    "npm pack @testdriverai/runner@${RUNNER_VERSION} 2>&1 | Write-Host",
+    "\$tarball = (Get-ChildItem 'C:\\\\Windows\\\\Temp\\\\testdriverai-runner-*.tgz' | Sort-Object LastWriteTime -Descending | Select-Object -First 1).FullName",
+    "Write-Host \"Downloaded tarball: \$tarball\"",
+    "Write-Host 'Extracting runner...'",
+    "tar -xzf \$tarball -C 'C:\\\\Windows\\\\Temp'",
+    "Write-Host 'New runner version in package:'",
+    "Get-Content 'C:\\\\Windows\\\\Temp\\\\package\\\\package.json' | ConvertFrom-Json | Select-Object -ExpandProperty version",
+    "Write-Host 'Clearing old lib folder...'",
+    "Remove-Item 'C:\\\\testdriver\\\\sandbox-agent\\\\lib' -Recurse -Force -ErrorAction SilentlyContinue",
+    "Write-Host 'Copying files to sandbox-agent...'",
+    "xcopy 'C:\\\\Windows\\\\Temp\\\\package\\\\*' 'C:\\\\testdriver\\\\sandbox-agent\\\\' /E /Y /I",
+    "Write-Host 'Runner version after copy:'",
+    "Get-Content 'C:\\\\testdriver\\\\sandbox-agent\\\\package.json' | ConvertFrom-Json | Select-Object -ExpandProperty version",
+    "Remove-Item 'C:\\\\Windows\\\\Temp\\\\package' -Recurse -Force -ErrorAction SilentlyContinue",
+    "Remove-Item \$tarball -Force -ErrorAction SilentlyContinue",
+    "Set-Location 'C:\\\\testdriver\\\\sandbox-agent'",
+    "Write-Host 'Installing npm dependencies...'",
+    "npm install --omit=dev 2>&1 | Write-Host",
+    "Write-Host 'Restarting RunTestDriverAgent scheduled task...'",
+    "Start-ScheduledTask -TaskName RunTestDriverAgent -ErrorAction SilentlyContinue",
+    "Write-Host '=== Runner install complete (npm pack) ==='"
+  ]
+}
+PARAMS_EOF
+
+  INSTALL_CMD=$(aws ssm send-command \
+    --region "$AWS_REGION" \
+    --instance-ids "$INSTANCE_ID" \
+    --document-name "AWS-RunPowerShellScript" \
+    --parameters "file://$PARAMS_FILE" \
+    --timeout-seconds 180 \
+    --output json)
+  rm -f "$PARAMS_FILE"
+fi
+
+INSTALL_CMD_ID=$(jq -r '.Command.CommandId' <<<"$INSTALL_CMD")
+echo "Runner install command sent (Command ID: $INSTALL_CMD_ID)"
+
+# Wait for install to complete
+echo "Waiting for runner install to complete..."
+if aws ssm wait command-executed --region "$AWS_REGION" --command-id "$INSTALL_CMD_ID" --instance-id "$INSTANCE_ID" 2>/dev/null; then
+  echo "✓ Runner install succeeded"
+else
+  INSTALL_STATUS=$(aws ssm get-command-invocation \
+    --region "$AWS_REGION" \
+    --command-id "$INSTALL_CMD_ID" \
+    --instance-id "$INSTANCE_ID" \
+    --output json 2>/dev/null || echo '{}')
+  echo "⚠ Runner install status: $(jq -r '.Status // "Unknown"' <<<"$INSTALL_STATUS")"
+  echo "Output: $(jq -r '.StandardOutputContent // "No output"' <<<"$INSTALL_STATUS" | head -20)"
+  echo "Errors: $(jq -r '.StandardErrorContent // "No errors"' <<<"$INSTALL_STATUS" | head -10)"
+fi
+
 echo "Getting Public IP..."
 
 # --- 5) Get instance Public IP ---