0agent 1.0.55 → 1.0.56

package/dist/daemon.mjs

@@ -2599,7 +2599,7 @@ var init_GUICapability = __esm({
       description = "Automate desktop GUI \u2014 click, type, screenshot, hotkeys, find text on screen.";
       toolDefinition = {
         name: "gui_automation",
-        description: 'Automate desktop GUI interactions. Take screenshots to see the current screen state, click on buttons/links/fields, type text, press keyboard shortcuts, scroll, open apps. IMPORTANT: Limit screenshots to at most 3 per task \u2014 avoid re-screenshotting if you already know the layout. Prefer targeted actions (click, find_and_click, hotkey) over repeated screenshots. Use get_cursor_pos to check cursor position without a full screenshot. To open a website, ALWAYS use action="open_url" \u2014 never open_app + new tab, which creates duplicate windows.',
+        description: "Desktop GUI automation \u2014 ONLY use for tasks that explicitly require controlling the screen. DO NOT use for coding, research, file edits, or any task that does not need the desktop UI. Actions: click, type, hotkey, scroll, find_and_click, open_url, open_app. screenshot: use sparingly \u2014 only when you cannot proceed without seeing the current screen. Max 2 per task. To open a website use open_url (reuses existing browser tab, never opens duplicate windows).",
         input_schema: {
           type: "object",
           properties: {
@@ -3497,12 +3497,14 @@ content = element.text if element else page.get_all_text()` : `content = page.ge
           `- Use relative paths from the working directory`,
           `- Be concise in your final response: state what was done and where to find it`,
           ``,
-          `GUI Automation (gui_automation tool):`,
-          `- ALWAYS call gui_automation({action:"screenshot"}) first to see what is on screen`,
-          `- Use the OCR output to find element coordinates before clicking`,
-          `- After clicking or typing, take another screenshot to confirm the result`,
+          `GUI Automation (gui_automation tool) \u2014 ONLY use when the task explicitly requires controlling the desktop UI:`,
+          `- DO NOT take screenshots for general tasks, coding, research, or anything that doesn't need the screen`,
+          `- Only screenshot when you genuinely cannot proceed without seeing the current screen state`,
+          `- Prefer find_and_click, hotkey, open_url, and type over repeated screenshots`,
+          `- Max 2 screenshots per task \u2014 if you've already seen the screen, act on that knowledge`,
           `- Use find_and_click to click on text by name rather than guessing coordinates`,
           `- Use hotkey for keyboard shortcuts: "cmd+c", "ctrl+v", "alt+tab", "cmd+space"`,
+          `- To open a website: use open_url \u2014 it reuses the existing browser tab`,
           ...hasMemory ? [
             ``,
             `Memory (CRITICAL \u2014 write EVERYTHING you learn):`,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "0agent",
-  "version": "1.0.55",
+  "version": "1.0.56",
   "description": "A persistent, learning AI agent that runs on your machine. An agent that learns.",
   "private": false,
   "license": "Apache-2.0",