nothumanallowed 9.3.2 → 9.3.4

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "nothumanallowed",
-  "version": "9.3.2",
+  "version": "9.3.4",
   "description": "NotHumanAllowed — 38 AI agents + 58 tools + browser automation + web search. Streaming chat, headless Chrome CDP, multi-conversation, export. Gmail, Calendar, Drive, GitHub, Notion, Slack. Zero-dependency CLI.",
   "type": "module",
   "bin": {

package/src/constants.mjs

@@ -5,7 +5,7 @@ import { fileURLToPath } from 'url';
 const __filename = fileURLToPath(import.meta.url);
 const __dirname = path.dirname(__filename);
 
-export const VERSION = '9.3.2';
+export const VERSION = '9.3.4';
 export const BASE_URL = 'https://nothumanallowed.com/cli';
 export const API_BASE = 'https://nothumanallowed.com/api/v1';
 

package/src/services/browser-engine.mjs

@@ -532,12 +532,17 @@ async function launchBrowser() {
 
   const port = await findFreePort();
 
+  // Realistic user agent to avoid bot detection on major sites
+  const UA = 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36';
+
   const chromeArgs = [
     `--remote-debugging-port=${port}`,
     '--headless=new',
     `--user-data-dir=${userDataDir}`,
+    `--user-agent=${UA}`,
     '--no-first-run',
     '--no-default-browser-check',
+    '--disable-blink-features=AutomationControlled',
     '--disable-background-networking',
     '--disable-background-timer-throttling',
     '--disable-backgrounding-occluded-windows',
@@ -603,6 +608,16 @@ async function launchBrowser() {
   await ws.send('Network.enable');
   await ws.send('DOM.enable');
 
+  // Anti-detection: remove navigator.webdriver flag and other headless indicators
+  await ws.send('Page.addScriptToEvaluateOnNewDocument', {
+    source: `
+      Object.defineProperty(navigator, 'webdriver', { get: () => false });
+      Object.defineProperty(navigator, 'languages', { get: () => ['it-IT', 'it', 'en-US', 'en'] });
+      Object.defineProperty(navigator, 'plugins', { get: () => [1, 2, 3, 4, 5] });
+      window.chrome = { runtime: {} };
+    `,
+  });
+
   _browser = { process: proc, ws, wsUrl, userDataDir, port };
 
   // Cleanup on process exit
@@ -756,10 +771,11 @@ export async function browserScreenshot(options = {}) {
 }
 
 /**
- * Click an element by CSS selector or coordinates.
+ * Click an element by CSS selector, visible text, or coordinates.
  *
  * @param {object} target
  * @param {string} [target.selector] - CSS selector to click
+ * @param {string} [target.text] - Visible text of a button/link/element to click (case-insensitive partial match)
  * @param {number} [target.x] - X coordinate
  * @param {number} [target.y] - Y coordinate
  * @returns {Promise<{ clicked: boolean, selector?: string }>}
@@ -768,9 +784,44 @@ export async function browserClick(target) {
   const browser = await getBrowser();
 
   let x, y;
+  let label = '';
+
+  if (target.text) {
+    // Find element by visible text — searches buttons, links, and all clickable elements
+    const searchText = target.text;
+    const result = await browser.ws.send('Runtime.evaluate', {
+      expression: `(() => {
+        const searchText = ${JSON.stringify(searchText)}.toLowerCase();
+        // Search in priority order: buttons, links, then any visible element
+        const candidates = [
+          ...document.querySelectorAll('button, [role="button"], input[type="submit"], input[type="button"]'),
+          ...document.querySelectorAll('a'),
+          ...document.querySelectorAll('[onclick], [tabindex]'),
+        ];
+        for (const el of candidates) {
+          const text = (el.textContent || el.value || el.getAttribute('aria-label') || '').trim();
+          if (text.toLowerCase().includes(searchText) && el.offsetHeight > 0 && el.offsetWidth > 0) {
+            const rect = el.getBoundingClientRect();
+            if (rect.width > 0 && rect.height > 0) {
+              return { x: rect.x + rect.width / 2, y: rect.y + rect.height / 2, tag: el.tagName, text: text.slice(0, 60) };
+            }
+          }
+        }
+        return null;
+      })()`,
+      returnByValue: true,
+    });
 
-  if (target.selector) {
-    // Find element and get its center point
+    const info = result.result?.value;
+    if (!info) {
+      return { error: true, message: `No clickable element found with text "${searchText}". Try browser_extract to see visible buttons.` };
+    }
+
+    x = Math.round(info.x);
+    y = Math.round(info.y);
+    label = `"${info.text}" (${info.tag})`;
+  } else if (target.selector) {
+    // Find element by CSS selector and get its center point
     const result = await browser.ws.send('Runtime.evaluate', {
       expression: `(() => {
         const el = document.querySelector(${JSON.stringify(target.selector)});
@@ -789,11 +840,13 @@ export async function browserClick(target) {
 
     x = Math.round(info.x);
     y = Math.round(info.y);
+    label = target.selector;
   } else if (target.x !== undefined && target.y !== undefined) {
     x = target.x;
     y = target.y;
+    label = `(${x}, ${y})`;
   } else {
-    return { error: true, message: 'Provide a CSS selector or x/y coordinates' };
+    return { error: true, message: 'Provide text, CSS selector, or x/y coordinates' };
   }
 
   // Dispatch mouse events: move → down → up (simulates real click)
@@ -807,10 +860,25 @@ export async function browserClick(target) {
     type: 'mouseReleased', x, y, button: 'left', clickCount: 1,
   });
 
-  // Brief wait for any navigation or JS handlers
-  await new Promise(r => setTimeout(r, 300));
+  // Wait for any navigation or JS handlers triggered by the click
+  await new Promise(r => setTimeout(r, 500));
+
+  // Check if the click triggered a navigation — wait for it
+  try {
+    const navCheck = await browser.ws.send('Runtime.evaluate', {
+      expression: 'document.readyState',
+      returnByValue: true,
+    });
+    if (navCheck.result?.value === 'loading') {
+      // Page is navigating — wait for load
+      try {
+        await browser.ws.waitForEvent('Page.loadEventFired', 10000);
+        await new Promise(r => setTimeout(r, 500));
+      } catch { /* timeout ok */ }
+    }
+  } catch { /* evaluation failed during navigation, that's fine */ }
 
-  return { error: false, clicked: true, x, y, selector: target.selector || `(${x}, ${y})` };
+  return { error: false, clicked: true, x, y, selector: label || target.selector || `(${x}, ${y})` };
 }
 
 /**

package/src/services/tool-executor.mjs

@@ -270,8 +270,8 @@ TOOLS:
 47. web_search(query: string, deep?: boolean)
     Search the web using DuckDuckGo. Returns titles, URLs, and snippets.
     Set deep=true to also fetch and extract the top 3 pages' full content (slower but more detailed).
-    Use this when the user asks about current events, recent news, or needs up-to-date information
-    that you don't have in your training data.
+    ALWAYS use this for ANY web search request ("search for X", "find X", "look up X", "cerca X").
+    Do NOT open Google/Bing in the browser for searches — use this tool instead. It's faster and never gets blocked.
 
 48. fetch_url(url: string)
     Fetch a web page and extract its text content. SSRF-protected (blocks private IPs, localhost).
@@ -284,16 +284,19 @@ TOOLS:
     Open a URL in a headless Chrome browser. Launches Chrome automatically on first use.
     SSRF-protected (blocks private IPs, localhost). Renders JavaScript, SPAs, and dynamic pages.
     Use this when you need to interact with a page (click, type, screenshot) or when fetch_url fails on JS-rendered content.
+    WARNING: Do NOT use this to search the web — use web_search instead. Google/Bing block automated browsers with CAPTCHAs.
 
 50. browser_screenshot(fullPage?: boolean, format?: "png"|"jpeg"|"webp", quality?: number, saveTo?: string)
     Capture a screenshot of the current browser page.
     fullPage=true captures the entire scrollable page. saveTo saves to a file path (e.g. "~/screenshot.png").
     Returns base64-encoded image data. Use after browser_open to see what's on screen.
 
-51. browser_click(selector?: string, x?: number, y?: number)
-    Click an element on the page by CSS selector or x/y coordinates.
-    Examples: selector="#submit-btn", selector="a.nav-link", selector="button:nth-of-type(2)"
-    Use coordinates for precise clicking when selector doesn't work.
+51. browser_click(text?: string, selector?: string, x?: number, y?: number)
+    Click an element on the page by visible text, CSS selector, or x/y coordinates.
+    PREFERRED: use text="Rifiuta tutto" or text="Submit" to click buttons/links by their visible label (case-insensitive partial match).
+    CSS selector: selector="#submit-btn", selector="a.nav-link"
+    Coordinates: x=500, y=300 for precise clicking.
+    Always try text first — it works for buttons, links, and any clickable element regardless of CSS structure.
 
 52. browser_type(text: string, selector?: string, clear?: boolean, delay?: number)
     Type text into an input field. If selector is provided, clicks the element first to focus it.
@@ -328,6 +331,7 @@ TOOLS:
     Close the browser. Frees resources. Browser auto-closes when NHA exits.
 
 RULES:
+- CRITICAL: For web searches ("search for X", "find X online", "look up X"), ALWAYS use web_search — NEVER open Google/Bing/DuckDuckGo in the browser. web_search is faster, more reliable, and doesn't get blocked by CAPTCHAs. Only use browser_open for interacting with specific websites (filling forms, clicking buttons, taking screenshots of specific pages).
 - For search/read operations, execute immediately and present results conversationally.
 - For write/send/delete operations (gmail_send, gmail_reply, gmail_delete, calendar_create, calendar_move, calendar_update, contact_delete, task_done, notify_remind), DESCRIBE what you're about to do and include the JSON block so the system can ask the user for confirmation.
 - For schedule_meeting and schedule_draft_email, execute immediately — these are read operations that suggest slots.
@@ -1129,6 +1133,7 @@ export async function executeTool(action, params, config) {
       if (!be.isBrowserRunning()) return 'No browser open. Use browser_open first.';
 
       const result = await be.browserClick({
+        text: params.text,
         selector: params.selector,
         x: params.x,
         y: params.y,