assistme 0.1.7 → 0.1.9

package/dist/index.js

@@ -432,9 +432,7 @@ var BrowserController = class {
     const available = await this.isAvailable();
     if (!available) {
       throw new Error(
-        `Cannot connect to browser on port ${this.debugPort}. Please start Chrome with: --remote-debugging-port=9222
-macOS: open -a 'Google Chrome' --args --remote-debugging-port=9222
-Linux: google-chrome --remote-debugging-port=9222`
+        `Cannot connect to browser on port ${this.debugPort}. Chrome remote debugging is not reachable. Please ensure Chrome is running with remote debugging enabled.`
       );
     }
     const tabs = await this.getTabs();
@@ -855,14 +853,24 @@ function findChromePath() {
 function isChromeRunning() {
   try {
     if (platform() === "win32") {
-      const out = execSync('tasklist /FI "IMAGENAME eq chrome.exe" /NH', {
+      const out2 = execSync('tasklist /FI "IMAGENAME eq chrome.exe" /NH', {
         encoding: "utf-8",
         stdio: ["pipe", "pipe", "pipe"]
       });
-      return out.includes("chrome.exe");
+      return out2.includes("chrome.exe");
     }
-    execSync("pgrep -f chrome", { stdio: ["pipe", "pipe", "pipe"] });
-    return true;
+    if (platform() === "darwin") {
+      const out2 = execSync('pgrep -f "Google Chrome.app/Contents/MacOS/Google Chrome"', {
+        encoding: "utf-8",
+        stdio: ["pipe", "pipe", "pipe"]
+      });
+      return out2.trim().length > 0;
+    }
+    const out = execSync("pgrep -f '(chrome|chromium)' 2>/dev/null || true", {
+      encoding: "utf-8",
+      stdio: ["pipe", "pipe", "pipe"]
+    });
+    return out.trim().length > 0;
   } catch {
     return false;
   }
@@ -908,52 +916,97 @@ async function killChromeGracefully() {
   await new Promise((r) => setTimeout(r, 1e3));
 }
 function spawnChrome(chromePath, port) {
-  const os = platform();
   const cdpFlag = `--remote-debugging-port=${port}`;
-  if (os === "darwin") {
-    const appName = chromePath.includes("Chromium") ? "Chromium" : chromePath.includes("Canary") ? "Google Chrome Canary" : "Google Chrome";
-    spawn("open", ["-a", appName, "--args", cdpFlag, "--restore-last-session"], {
-      detached: true,
-      stdio: "ignore"
-    }).unref();
-  } else {
-    spawn(chromePath, [cdpFlag, "--restore-last-session"], {
-      detached: true,
-      stdio: "ignore"
-    }).unref();
-  }
+  log.debug(`Spawning Chrome: ${chromePath} ${cdpFlag} --restore-last-session`);
+  const child = spawn(chromePath, [cdpFlag, "--restore-last-session"], {
+    detached: true,
+    stdio: "ignore"
+  });
+  child.on("error", (err) => {
+    log.error(`Chrome spawn error: ${err.message}`);
+  });
+  child.unref();
+  return child;
 }
-async function waitForCDP(browser, timeoutMs = 15e3) {
+async function waitForCDP(browser, timeoutMs = 3e4) {
   const start = Date.now();
+  let attempts = 0;
   while (Date.now() - start < timeoutMs) {
-    if (await browser.isAvailable()) return true;
+    attempts++;
+    if (await browser.isAvailable()) {
+      log.debug(`CDP became reachable after ${attempts} attempts (${Date.now() - start}ms)`);
+      return true;
+    }
     await new Promise((r) => setTimeout(r, 500));
   }
+  log.debug(`CDP not reachable after ${attempts} attempts (${timeoutMs}ms timeout)`);
   return false;
 }
+async function isPortInUse(port) {
+  try {
+    const res = await fetch(`http://127.0.0.1:${port}/json/version`, {
+      signal: AbortSignal.timeout(1e3)
+    });
+    const body = await res.text();
+    return !body.includes("Chrome");
+  } catch {
+    return false;
+  }
+}
 async function ensureBrowserAvailable(port = 9222) {
   const browser = getBrowser(port);
   if (await browser.isAvailable()) {
+    log.debug("CDP already reachable \u2014 no launch needed");
     return { success: true, action: "already_available" };
   }
+  if (await isPortInUse(port)) {
+    log.debug(`Port ${port} is in use by a non-Chrome process`);
+    return {
+      success: false,
+      action: "port_conflict",
+      detail: `Port ${port} is already in use by another process. Try a different port or stop the conflicting process.`
+    };
+  }
   const chromePath = findChromePath();
   if (!chromePath) {
+    log.debug("Chrome binary not found on this system");
     return { success: false, action: "chrome_not_found" };
   }
+  log.debug(`Found Chrome at: ${chromePath}`);
   const running = isChromeRunning();
+  log.debug(`Chrome currently running: ${running}`);
   if (running) {
+    log.debug("Killing Chrome gracefully for restart with CDP...");
     await killChromeGracefully();
     spawnChrome(chromePath, port);
     if (await waitForCDP(browser)) {
       return { success: true, action: "restarted", chromePath };
     }
-    return { success: false, action: "launch_failed", chromePath };
+    log.debug("First CDP wait timed out after restart, retrying...");
+    if (await waitForCDP(browser, 15e3)) {
+      return { success: true, action: "restarted", chromePath };
+    }
+    return {
+      success: false,
+      action: "launch_failed",
+      chromePath,
+      detail: "Chrome was restarted but CDP did not become reachable within timeout."
+    };
   }
   spawnChrome(chromePath, port);
   if (await waitForCDP(browser)) {
     return { success: true, action: "launched", chromePath };
   }
-  return { success: false, action: "launch_failed", chromePath };
+  log.debug("First CDP wait timed out after launch, retrying...");
+  if (await waitForCDP(browser, 15e3)) {
+    return { success: true, action: "launched", chromePath };
+  }
+  return {
+    success: false,
+    action: "launch_failed",
+    chromePath,
+    detail: "Chrome was launched but CDP did not become reachable within timeout."
+  };
 }
 var browserInstance = null;
 function getBrowser(port = 9222) {
@@ -2164,8 +2217,17 @@ async function executeTool(name, input) {
     case "execute_command":
       return executeShell(input.command, input.cwd);
     // ── Browser (CDP) ───────────────────────────────────────
-    case "browser_connect":
+    case "browser_connect": {
+      if (!await browser.isAvailable()) {
+        const result = await ensureBrowserAvailable();
+        if (!result.success) {
+          throw new Error(
+            `Failed to auto-launch Chrome (${result.action}). Please ensure Google Chrome is installed.`
+          );
+        }
+      }
       return browser.connect(input.tab_index);
+    }
     case "browser_navigate":
       if (!browser.isConnected()) await browser.connect();
       return browser.navigate(input.url);
@@ -2341,7 +2403,7 @@ function createBrowserMcpServer() {
     tools: [
       tool(
         "browser_connect",
-        "Connect to the user's real Chrome browser via CDP. The user must have Chrome running with --remote-debugging-port=9222.",
+        "Connect to the user's real Chrome browser via CDP. Chrome will be auto-launched if not already running.",
         { tab_index: z.number().optional().describe("Tab index (default: 0)") },
         async (args) => callTool("browser_connect", args)
       ),
@@ -2459,18 +2521,14 @@ function createAgentToolsServer(deps) {
         "Store a memory about the user that persists across conversations. Use when you learn preferences, habits, or standing instructions.",
         {
           content: z.string().describe("What to remember (concise, factual statement)"),
-          category: z.string().optional().describe(
-            "Category: general, preference, instruction, context, skill_learned, fact"
-          ),
+          category: z.string().optional().describe("Category: general, preference, instruction, context, skill_learned, fact"),
           importance: z.number().optional().describe("Importance 1-10 (default: 5). Use 8+ for instructions"),
           tags: z.array(z.string()).optional().describe("Optional tags for searchability")
         },
         async (args) => {
           if (!memoryManager) {
             return {
-              content: [
-                { type: "text", text: "Memory manager not available." }
-              ]
+              content: [{ type: "text", text: "Memory manager not available." }]
             };
           }
           const mem = await memoryManager.remember(
@@ -2507,11 +2565,7 @@ function createAgentToolsServer(deps) {
               ]
             };
           }
-          const filePath = skillManager.create(
-            args.name,
-            args.description,
-            args.instructions
-          );
+          const filePath = skillManager.create(args.name, args.description, args.instructions);
           if (args.emoji) {
             const skill = skillManager.get(args.name);
             if (skill) {
@@ -2571,9 +2625,7 @@ ${args.instructions}
               args.description || existing.description,
               args.improved_instructions
             );
-            log.success(
-              `Self-improvement: overrode bundled skill "${args.name}"`
-            );
+            log.success(`Self-improvement: overrode bundled skill "${args.name}"`);
             return {
               content: [
                 {
@@ -2670,6 +2722,8 @@ KEY PRINCIPLE: You operate the user's real browser, not a headless sandbox. This
 - When you navigate to amazon.com, you see the user's logged-in Amazon
 - If a site needs login, ask the user to log in using browser_request_user_action
 - You are like a human assistant sitting at the user's computer
+- Chrome is automatically managed \u2014 just call browser_connect and it will auto-launch if needed
+- NEVER ask the user to manually start Chrome or run any terminal commands for browser setup
 
 Available capabilities:
 1. BROWSER CONTROL (user's real Chrome via CDP):
@@ -2731,10 +2785,7 @@ var TaskProcessor = class {
     let tokenUsage;
     try {
       await emitEvent(task.id, "status_change", { status: "running" });
-      let systemPrompt = BASE_SYSTEM_PROMPT.replace(
-        "{workspace_path}",
-        config.workspacePath
-      );
+      let systemPrompt = BASE_SYSTEM_PROMPT.replace("{workspace_path}", config.workspacePath);
       if (this.memoryManager) {
         try {
           const memoryPrompt = await this.memoryManager.buildMemoryPrompt();
@@ -2769,9 +2820,7 @@ var TaskProcessor = class {
         "Glob",
         "Grep",
         // Browser MCP tools
-        ...BROWSER_TOOL_NAMES.map(
-          (n) => `mcp__assistme-browser__${n}`
-        ),
+        ...BROWSER_TOOL_NAMES.map((n) => `mcp__assistme-browser__${n}`),
         // Agent MCP tools (memory, skills)
         "mcp__assistme-agent__memory_store",
         "mcp__assistme-agent__skill_create",
@@ -2830,9 +2879,7 @@ var TaskProcessor = class {
                   });
                 } else if (block.type === "thinking" && "thinking" in block) {
                   const thinkingText = block.thinking;
-                  log.debug(
-                    `Thinking: ${thinkingText.slice(0, 100)}...`
-                  );
+                  log.debug(`Thinking: ${thinkingText.slice(0, 100)}...`);
                   await emitEvent(task.id, "thinking", {
                     text: thinkingText
                   });
@@ -2871,14 +2918,11 @@ var TaskProcessor = class {
       } finally {
         clearTimeout(timeoutId);
       }
-      await withRetry(
-        () => completeTask(task.id, finalResponse, tokenUsage),
-        {
-          maxRetries: 2,
-          baseDelayMs: 300,
-          label: "completeTask"
-        }
-      );
+      await withRetry(() => completeTask(task.id, finalResponse, tokenUsage), {
+        maxRetries: 2,
+        baseDelayMs: 300,
+        label: "completeTask"
+      });
       await emitEvent(task.id, "status_change", { status: "completed" });
       log.success("Task completed.");
       if (this.memoryManager && finalResponse) {
@@ -2892,9 +2936,7 @@ var TaskProcessor = class {
                 tags: mem.tags,
                 sourceMessageId: taskIdRef
               });
-              log.info(
-                `Memory extracted: [${mem.category}] ${mem.content.slice(0, 60)}...`
-              );
+              log.info(`Memory extracted: [${mem.category}] ${mem.content.slice(0, 60)}...`);
             } catch {
             }
           }
@@ -2918,11 +2960,7 @@ var TaskProcessor = class {
             );
             return;
           }
-          const filePath = sm.create(
-            extracted.name,
-            extracted.description,
-            extracted.steps
-          );
+          const filePath = sm.create(extracted.name, extracted.description, extracted.steps);
           if (extracted.emoji) {
             const { writeFile: writeFile2 } = await import("fs/promises");
             const metaJson = JSON.stringify({
@@ -2948,19 +2986,10 @@ ${extracted.steps}
         for (const skillName of usedSkillNames) {
           const skill = sm.get(skillName);
           if (!skill) continue;
-          analyzeSkillImprovement(
-            skill.content,
-            task.prompt,
-            finalResponse,
-            realToolCalls
-          ).then(async (improvement) => {
+          analyzeSkillImprovement(skill.content, task.prompt, finalResponse, realToolCalls).then(async (improvement) => {
             if (!improvement) return;
             if (skill.source === "bundled") {
-              sm.create(
-                skillName,
-                skill.description,
-                improvement.improved_steps
-              );
+              sm.create(skillName, skill.description, improvement.improved_steps);
             } else {
               sm.update(skillName, improvement.improved_steps);
             }
@@ -3216,11 +3245,20 @@ program.command("start", { isDefault: true }).description("Start the agent and l
       case "chrome_not_found":
         launchSpinner.fail("Chrome not found on this system");
         log.info("Please install Google Chrome and try again.");
-        log.info('Or run "assistme browser setup" for manual instructions.');
+        break;
+      case "port_conflict":
+        launchSpinner.fail("Port 9222 is in use by another process");
+        log.info(launchResult.detail ?? "Stop the conflicting process or use a different port.");
         break;
       default:
         launchSpinner.fail("Failed to start Chrome with remote debugging");
-        log.info('Run "assistme browser setup" for manual setup instructions.');
+        if (launchResult.detail) {
+          log.info(launchResult.detail);
+        }
+        if (launchResult.chromePath) {
+          log.info(`Chrome binary: ${launchResult.chromePath}`);
+        }
+        log.info("Browser will be auto-launched when the first task needs it.");
         break;
     }
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "assistme",
-  "version": "0.1.7",
+  "version": "0.1.9",
   "description": "AssistMe CLI Agent - AI-powered assistant that controls your real browser",
   "type": "module",
   "main": "dist/index.js",

package/src/agent/mcp-servers.ts

@@ -49,7 +49,7 @@ export function createBrowserMcpServer(): McpSdkServerConfigWithInstance {
     tools: [
       tool(
         "browser_connect",
-        "Connect to the user's real Chrome browser via CDP. The user must have Chrome running with --remote-debugging-port=9222.",
+        "Connect to the user's real Chrome browser via CDP. Chrome will be auto-launched if not already running.",
         { tab_index: z.number().optional().describe("Tab index (default: 0)") },
         async (args) => callTool("browser_connect", args)
       ),
@@ -126,11 +126,8 @@ export function createBrowserMcpServer(): McpSdkServerConfigWithInstance {
         { expression: z.string().describe("JavaScript expression to evaluate") },
         async (args) => callTool("browser_evaluate", args)
       ),
-      tool(
-        "browser_list_tabs",
-        "List all open tabs in the user's browser.",
-        {},
-        async () => callTool("browser_list_tabs", {})
+      tool("browser_list_tabs", "List all open tabs in the user's browser.", {}, async () =>
+        callTool("browser_list_tabs", {})
       ),
       tool(
         "browser_switch_tab",
@@ -148,13 +145,8 @@ export function createBrowserMcpServer(): McpSdkServerConfigWithInstance {
         "browser_request_user_action",
         "Request the user to perform an action in their browser (login, CAPTCHA, 2FA, etc.).",
         {
-          message: z
-            .string()
-            .describe("Clear description of what the user needs to do"),
-          wait_seconds: z
-            .number()
-            .optional()
-            .describe("How long to wait (default: 60)"),
+          message: z.string().describe("Clear description of what the user needs to do"),
+          wait_seconds: z.number().optional().describe("How long to wait (default: 60)"),
         },
         async (args) => callTool("browser_request_user_action", args)
       ),
@@ -170,9 +162,7 @@ export interface AgentToolsDeps {
   taskId: string;
 }
 
-export function createAgentToolsServer(
-  deps: AgentToolsDeps
-): McpSdkServerConfigWithInstance {
+export function createAgentToolsServer(deps: AgentToolsDeps): McpSdkServerConfigWithInstance {
   const { memoryManager, skillManager, taskId } = deps;
 
   return createSdkMcpServer({
@@ -183,30 +173,21 @@ export function createAgentToolsServer(
         "memory_store",
         "Store a memory about the user that persists across conversations. Use when you learn preferences, habits, or standing instructions.",
         {
-          content: z
-            .string()
-            .describe("What to remember (concise, factual statement)"),
+          content: z.string().describe("What to remember (concise, factual statement)"),
           category: z
             .string()
             .optional()
-            .describe(
-              "Category: general, preference, instruction, context, skill_learned, fact"
-            ),
+            .describe("Category: general, preference, instruction, context, skill_learned, fact"),
           importance: z
             .number()
             .optional()
             .describe("Importance 1-10 (default: 5). Use 8+ for instructions"),
-          tags: z
-            .array(z.string())
-            .optional()
-            .describe("Optional tags for searchability"),
+          tags: z.array(z.string()).optional().describe("Optional tags for searchability"),
         },
         async (args) => {
           if (!memoryManager) {
             return {
-              content: [
-                { type: "text", text: "Memory manager not available." },
-              ],
+              content: [{ type: "text", text: "Memory manager not available." }],
             };
           }
           const mem = await memoryManager.remember(
@@ -226,19 +207,10 @@ export function createAgentToolsServer(
         "skill_create",
         "Create a new reusable skill from a workflow you just executed. Write generic, reusable instructions with placeholders like {product}, {query}.",
         {
-          name: z
-            .string()
-            .describe("Skill name in kebab-case, e.g. 'flight-booking'"),
-          description: z
-            .string()
-            .describe("One-line description of what this skill does"),
-          instructions: z
-            .string()
-            .describe("Markdown step-by-step instructions"),
-          emoji: z
-            .string()
-            .optional()
-            .describe("Single emoji representing this skill"),
+          name: z.string().describe("Skill name in kebab-case, e.g. 'flight-booking'"),
+          description: z.string().describe("One-line description of what this skill does"),
+          instructions: z.string().describe("Markdown step-by-step instructions"),
+          emoji: z.string().optional().describe("Single emoji representing this skill"),
         },
         async (args) => {
           // Check for duplicates
@@ -254,11 +226,7 @@ export function createAgentToolsServer(
             };
           }
 
-          const filePath = skillManager.create(
-            args.name,
-            args.description,
-            args.instructions
-          );
+          const filePath = skillManager.create(args.name, args.description, args.instructions);
 
           // Add emoji metadata if provided
           if (args.emoji) {
@@ -293,10 +261,7 @@ export function createAgentToolsServer(
           improved_instructions: z
             .string()
             .describe("Full updated markdown instructions (not a diff)"),
-          description: z
-            .string()
-            .optional()
-            .describe("Updated description (optional)"),
+          description: z.string().optional().describe("Updated description (optional)"),
         },
         async (args) => {
           const existing = skillManager.get(args.name);
@@ -321,9 +286,7 @@ export function createAgentToolsServer(
               args.description || existing.description,
               args.improved_instructions
             );
-            log.success(
-              `Self-improvement: overrode bundled skill "${args.name}"`
-            );
+            log.success(`Self-improvement: overrode bundled skill "${args.name}"`);
             return {
               content: [
                 {

package/src/agent/processor.ts

@@ -39,6 +39,8 @@ KEY PRINCIPLE: You operate the user's real browser, not a headless sandbox. This
 - When you navigate to amazon.com, you see the user's logged-in Amazon
 - If a site needs login, ask the user to log in using browser_request_user_action
 - You are like a human assistant sitting at the user's computer
+- Chrome is automatically managed — just call browser_connect and it will auto-launch if needed
+- NEVER ask the user to manually start Chrome or run any terminal commands for browser setup
 
 Available capabilities:
 1. BROWSER CONTROL (user's real Chrome via CDP):
@@ -97,7 +99,9 @@ export class TaskProcessor {
     resetEventSequence();
 
     // Wall-clock timeout for the entire task (default: 10 minutes)
-    const taskTimeoutMs = ((config as unknown as Record<string, unknown>).taskTimeoutMinutes as number || 10) * 60_000;
+    const taskTimeoutMs =
+      (((config as unknown as Record<string, unknown>).taskTimeoutMinutes as number) || 10) *
+      60_000;
 
     // Set correlation ID for this task's log messages
     newCorrelationId();
@@ -113,10 +117,7 @@ export class TaskProcessor {
       await emitEvent(task.id, "status_change", { status: "running" });
 
       // Build system prompt with memories + skills
-      let systemPrompt = BASE_SYSTEM_PROMPT.replace(
-        "{workspace_path}",
-        config.workspacePath
-      );
+      let systemPrompt = BASE_SYSTEM_PROMPT.replace("{workspace_path}", config.workspacePath);
 
       // Inject memories
       if (this.memoryManager) {
@@ -163,9 +164,7 @@ export class TaskProcessor {
         "Glob",
         "Grep",
         // Browser MCP tools
-        ...BROWSER_TOOL_NAMES.map(
-          (n) => `mcp__assistme-browser__${n}`
-        ),
+        ...BROWSER_TOOL_NAMES.map((n) => `mcp__assistme-browser__${n}`),
         // Agent MCP tools (memory, skills)
         "mcp__assistme-agent__memory_store",
         "mcp__assistme-agent__skill_create",
@@ -233,9 +232,7 @@ export class TaskProcessor {
                   });
                 } else if (block.type === "thinking" && "thinking" in block) {
                   const thinkingText = (block as unknown as { thinking: string }).thinking;
-                  log.debug(
-                    `Thinking: ${thinkingText.slice(0, 100)}...`
-                  );
+                  log.debug(`Thinking: ${thinkingText.slice(0, 100)}...`);
                   await emitEvent(task.id, "thinking", {
                     text: thinkingText,
                   });
@@ -283,14 +280,11 @@ export class TaskProcessor {
       }
 
       // Complete the task (with retry for transient DB failures)
-      await withRetry(
-        () => completeTask(task.id, finalResponse, tokenUsage),
-        {
-          maxRetries: 2,
-          baseDelayMs: 300,
-          label: "completeTask",
-        }
-      );
+      await withRetry(() => completeTask(task.id, finalResponse, tokenUsage), {
+        maxRetries: 2,
+        baseDelayMs: 300,
+        label: "completeTask",
+      });
       await emitEvent(task.id, "status_change", { status: "completed" });
       log.success("Task completed.");
 
@@ -309,9 +303,7 @@ export class TaskProcessor {
                   tags: mem.tags,
                   sourceMessageId: taskIdRef,
                 });
-                log.info(
-                  `Memory extracted: [${mem.category}] ${mem.content.slice(0, 60)}...`
-                );
+                log.info(`Memory extracted: [${mem.category}] ${mem.content.slice(0, 60)}...`);
               } catch {
                 // Non-critical — skip individual memory failures
               }
@@ -326,9 +318,7 @@ export class TaskProcessor {
       // Auto-extract skills from multi-step workflows
       const realToolCalls = toolCallRecords.filter(
         (tc) =>
-          tc.name !== "memory_store" &&
-          tc.name !== "skill_create" &&
-          tc.name !== "skill_improve"
+          tc.name !== "memory_store" && tc.name !== "skill_create" && tc.name !== "skill_improve"
       );
 
       if (realToolCalls.length >= 3 && finalResponse) {
@@ -347,11 +337,7 @@ export class TaskProcessor {
               return;
             }
 
-            const filePath = sm.create(
-              extracted.name,
-              extracted.description,
-              extracted.steps
-            );
+            const filePath = sm.create(extracted.name, extracted.description, extracted.steps);
 
             if (extracted.emoji) {
               const { writeFile } = await import("fs/promises");
@@ -373,21 +359,12 @@ export class TaskProcessor {
           const skill = sm.get(skillName);
           if (!skill) continue;
 
-          analyzeSkillImprovement(
-            skill.content,
-            task.prompt,
-            finalResponse,
-            realToolCalls
-          )
+          analyzeSkillImprovement(skill.content, task.prompt, finalResponse, realToolCalls)
             .then(async (improvement) => {
               if (!improvement) return;
 
               if (skill.source === "bundled") {
-                sm.create(
-                  skillName,
-                  skill.description,
-                  improvement.improved_steps
-                );
+                sm.create(skillName, skill.description, improvement.improved_steps);
               } else {
                 sm.update(skillName, improvement.improved_steps);
               }

package/src/index.ts

@@ -332,11 +332,20 @@ program
         case "chrome_not_found":
           launchSpinner.fail("Chrome not found on this system");
           log.info("Please install Google Chrome and try again.");
-          log.info('Or run "assistme browser setup" for manual instructions.');
+          break;
+        case "port_conflict":
+          launchSpinner.fail("Port 9222 is in use by another process");
+          log.info(launchResult.detail ?? "Stop the conflicting process or use a different port.");
           break;
         default:
           launchSpinner.fail("Failed to start Chrome with remote debugging");
-          log.info('Run "assistme browser setup" for manual setup instructions.');
+          if (launchResult.detail) {
+            log.info(launchResult.detail);
+          }
+          if (launchResult.chromePath) {
+            log.info(`Chrome binary: ${launchResult.chromePath}`);
+          }
+          log.info("Browser will be auto-launched when the first task needs it.");
           break;
       }
     }

package/src/tools/browser.ts

@@ -15,9 +15,10 @@
  */
 
 import { WebSocket } from "ws";
-import { execSync, spawn } from "node:child_process";
+import { execSync, spawn, type ChildProcess } from "node:child_process";
 import { platform } from "node:os";
 import { existsSync } from "node:fs";
+import { log } from "../utils/logger.js";
 
 interface CDPTab {
   id: string;
@@ -89,9 +90,8 @@ export class BrowserController {
     if (!available) {
       throw new Error(
         `Cannot connect to browser on port ${this.debugPort}. ` +
-          "Please start Chrome with: --remote-debugging-port=9222\n" +
-          "macOS: open -a 'Google Chrome' --args --remote-debugging-port=9222\n" +
-          "Linux: google-chrome --remote-debugging-port=9222"
+          "Chrome remote debugging is not reachable. " +
+          "Please ensure Chrome is running with remote debugging enabled."
       );
     }
 
@@ -606,9 +606,21 @@ export function isChromeRunning(): boolean {
       });
       return out.includes("chrome.exe");
     }
-    // macOS and Linux — pgrep automatically excludes its own process
-    execSync("pgrep -f chrome", { stdio: ["pipe", "pipe", "pipe"] });
-    return true;
+    if (platform() === "darwin") {
+      // Match the main Chrome process (not helper/renderer sub-processes).
+      // No trailing $ — the process command line includes flags after the binary.
+      const out = execSync('pgrep -f "Google Chrome.app/Contents/MacOS/Google Chrome"', {
+        encoding: "utf-8",
+        stdio: ["pipe", "pipe", "pipe"],
+      });
+      return out.trim().length > 0;
+    }
+    // Linux — match common chrome binary names
+    const out = execSync("pgrep -f '(chrome|chromium)' 2>/dev/null || true", {
+      encoding: "utf-8",
+      stdio: ["pipe", "pipe", "pipe"],
+    });
+    return out.trim().length > 0;
   } catch {
     return false;
   }
@@ -667,93 +679,164 @@ async function killChromeGracefully(): Promise<void> {
 
 /**
  * Spawn Chrome with the remote-debugging-port flag.
+ * Returns the child process so callers can detect early failures.
  */
-function spawnChrome(chromePath: string, port: number): void {
-  const os = platform();
+function spawnChrome(chromePath: string, port: number): ChildProcess {
   const cdpFlag = `--remote-debugging-port=${port}`;
 
-  if (os === "darwin") {
-    // Determine app name from binary path
-    const appName = chromePath.includes("Chromium")
-      ? "Chromium"
-      : chromePath.includes("Canary")
-        ? "Google Chrome Canary"
-        : "Google Chrome";
-    spawn("open", ["-a", appName, "--args", cdpFlag, "--restore-last-session"], {
-      detached: true,
-      stdio: "ignore",
-    }).unref();
-  } else {
-    spawn(chromePath, [cdpFlag, "--restore-last-session"], {
-      detached: true,
-      stdio: "ignore",
-    }).unref();
-  }
+  // Always invoke the Chrome binary directly rather than `open -a`.
+  // On macOS, `open -a` silently ignores --args when Chrome is already
+  // running, which would cause CDP to never be enabled.
+  log.debug(`Spawning Chrome: ${chromePath} ${cdpFlag} --restore-last-session`);
+  const child = spawn(chromePath, [cdpFlag, "--restore-last-session"], {
+    detached: true,
+    stdio: "ignore",
+  });
+
+  child.on("error", (err) => {
+    log.error(`Chrome spawn error: ${err.message}`);
+  });
+
+  child.unref();
+  return child;
 }
 
 /**
  * Wait for CDP to become reachable.
  */
-async function waitForCDP(browser: BrowserController, timeoutMs = 15000): Promise<boolean> {
+async function waitForCDP(browser: BrowserController, timeoutMs = 30000): Promise<boolean> {
   const start = Date.now();
+  let attempts = 0;
   while (Date.now() - start < timeoutMs) {
-    if (await browser.isAvailable()) return true;
+    attempts++;
+    if (await browser.isAvailable()) {
+      log.debug(`CDP became reachable after ${attempts} attempts (${Date.now() - start}ms)`);
+      return true;
+    }
     await new Promise((r) => setTimeout(r, 500));
   }
+  log.debug(`CDP not reachable after ${attempts} attempts (${timeoutMs}ms timeout)`);
   return false;
 }
 
+/**
+ * Check if a port is already in use by another process (not Chrome CDP).
+ */
+async function isPortInUse(port: number): Promise<boolean> {
+  try {
+    const res = await fetch(`http://127.0.0.1:${port}/json/version`, {
+      signal: AbortSignal.timeout(1000),
+    });
+    // If we get a response but it's not Chrome, the port is occupied
+    const body = await res.text();
+    return !body.includes("Chrome");
+  } catch {
+    // Connection refused → port is free
+    return false;
+  }
+}
+
 /**
  * Result of an auto-launch attempt.
  */
 export interface AutoLaunchResult {
   success: boolean;
-  action: "already_available" | "launched" | "restarted" | "chrome_not_found" | "launch_failed";
+  action:
+    | "already_available"
+    | "launched"
+    | "restarted"
+    | "chrome_not_found"
+    | "launch_failed"
+    | "port_conflict";
   chromePath?: string;
+  detail?: string;
 }
 
 /**
  * Ensure Chrome is running with CDP enabled.
  *
  * 1. Already listening on the port → return immediately.
- * 2. Chrome not running → launch with --remote-debugging-port.
- * 3. Chrome running without CDP → graceful quit, then relaunch with CDP.
+ * 2. Port occupied by non-Chrome process → report conflict.
+ * 3. Chrome not running → launch with --remote-debugging-port.
+ * 4. Chrome running without CDP → graceful quit, then relaunch with CDP.
  *    Chrome's session restore brings back all tabs.
+ *
+ * On launch failure, retries once with a longer wait.
  */
 export async function ensureBrowserAvailable(port = 9222): Promise<AutoLaunchResult> {
   const browser = getBrowser(port);
 
   // Case 1: CDP already reachable
   if (await browser.isAvailable()) {
+    log.debug("CDP already reachable — no launch needed");
     return { success: true, action: "already_available" };
   }
 
+  // Case 2: Port occupied by something else
+  if (await isPortInUse(port)) {
+    log.debug(`Port ${port} is in use by a non-Chrome process`);
+    return {
+      success: false,
+      action: "port_conflict",
+      detail: `Port ${port} is already in use by another process. Try a different port or stop the conflicting process.`,
+    };
+  }
+
   // Find Chrome binary
   const chromePath = findChromePath();
   if (!chromePath) {
+    log.debug("Chrome binary not found on this system");
     return { success: false, action: "chrome_not_found" };
   }
 
+  log.debug(`Found Chrome at: ${chromePath}`);
+
   const running = isChromeRunning();
+  log.debug(`Chrome currently running: ${running}`);
 
-  // Case 2: Chrome running without CDP → restart
+  // Case 3: Chrome running without CDP → restart
   if (running) {
+    log.debug("Killing Chrome gracefully for restart with CDP...");
     await killChromeGracefully();
     spawnChrome(chromePath, port);
 
     if (await waitForCDP(browser)) {
       return { success: true, action: "restarted", chromePath };
     }
-    return { success: false, action: "launch_failed", chromePath };
+
+    // Retry once — Chrome can be slow to start (extensions, session restore)
+    log.debug("First CDP wait timed out after restart, retrying...");
+    if (await waitForCDP(browser, 15000)) {
+      return { success: true, action: "restarted", chromePath };
+    }
+
+    return {
+      success: false,
+      action: "launch_failed",
+      chromePath,
+      detail: "Chrome was restarted but CDP did not become reachable within timeout.",
+    };
   }
 
-  // Case 3: Chrome not running → launch
+  // Case 4: Chrome not running → launch
   spawnChrome(chromePath, port);
 
   if (await waitForCDP(browser)) {
     return { success: true, action: "launched", chromePath };
   }
-  return { success: false, action: "launch_failed", chromePath };
+
+  // Retry once
+  log.debug("First CDP wait timed out after launch, retrying...");
+  if (await waitForCDP(browser, 15000)) {
+    return { success: true, action: "launched", chromePath };
+  }
+
+  return {
+    success: false,
+    action: "launch_failed",
+    chromePath,
+    detail: "Chrome was launched but CDP did not become reachable within timeout.",
+  };
 }
 
 // ── Singleton ───────────────────────────────────────────────────────

package/src/tools/index.ts

@@ -1,4 +1,4 @@
-import { getBrowser } from "./browser.js";
+import { getBrowser, ensureBrowserAvailable } from "./browser.js";
 import {
   readFileContent,
   writeFileContent,
@@ -99,13 +99,14 @@ export function getToolDefinitions(): ToolDefinition[] {
     {
       name: "browser_connect",
       description:
-        "Connect to the user's real Chrome browser via CDP. The user must have Chrome running with --remote-debugging-port=9222. This shares the user's actual browser session including all logins and cookies.",
+        "Connect to the user's real Chrome browser via CDP. Chrome will be auto-launched if not already running. This shares the user's actual browser session including all logins and cookies.",
       input_schema: {
         type: "object",
         properties: {
           tab_index: {
             type: "number",
-            description: "Tab index to connect to (default: 0, the first tab). Use browser_list_tabs to see available tabs.",
+            description:
+              "Tab index to connect to (default: 0, the first tab). Use browser_list_tabs to see available tabs.",
           },
         },
       },
@@ -143,7 +144,8 @@ export function getToolDefinitions(): ToolDefinition[] {
         properties: {
           selector: {
             type: "string",
-            description: "CSS selector of the element to click (e.g. '#submit-btn', 'a.nav-link', 'button:nth-of-type(2)')",
+            description:
+              "CSS selector of the element to click (e.g. '#submit-btn', 'a.nav-link', 'button:nth-of-type(2)')",
           },
         },
         required: ["selector"],
@@ -268,10 +270,7 @@ export function getToolDefinitions(): ToolDefinition[] {
   ];
 }
 
-export async function executeTool(
-  name: string,
-  input: Record<string, unknown>
-): Promise<string> {
+export async function executeTool(name: string, input: Record<string, unknown>): Promise<string> {
   const browser = getBrowser();
 
   switch (name) {
@@ -298,8 +297,19 @@ export async function executeTool(
       return executeShell(input.command as string, input.cwd as string | undefined);
 
     // ── Browser (CDP) ───────────────────────────────────────
-    case "browser_connect":
+    case "browser_connect": {
+      // Auto-launch Chrome if CDP is not reachable
+      if (!(await browser.isAvailable())) {
+        const result = await ensureBrowserAvailable();
+        if (!result.success) {
+          throw new Error(
+            `Failed to auto-launch Chrome (${result.action}). ` +
+              "Please ensure Google Chrome is installed."
+          );
+        }
+      }
       return browser.connect(input.tab_index as number | undefined);
+    }
     case "browser_navigate":
       if (!browser.isConnected()) await browser.connect();
       return browser.navigate(input.url as string);
@@ -314,9 +324,7 @@ export async function executeTool(
     case "browser_press_key":
       return browser.pressKey(input.key as string);
     case "browser_scroll":
-      return (input.direction as string) === "up"
-        ? browser.scrollUp()
-        : browser.scrollDown();
+      return (input.direction as string) === "up" ? browser.scrollUp() : browser.scrollDown();
     case "browser_get_elements":
       return browser.getInteractiveElements();
     case "browser_evaluate":