0agent 1.0.64 → 1.0.66

package/dist/daemon.mjs

@@ -3371,8 +3371,9 @@ var init_CapabilityRegistry = __esm({
           active.add("scrape_url");
           active.add("browser_open");
         }
-        if (/click|screenshot|ui|desktop|window|screen|gui|mouse|keyboard|open.*app|fill.*form|navigate.*browser|interact|automate|computer.*use/i.test(lower)) {
+        if (/click|screenshot|ui|desktop|window|screen|gui|mouse|keyboard|open.*app|fill.*form|navigate.*browser|interact|automate|computer.*use|whatsapp|telegram|youtube|music|play|send.*message/i.test(lower)) {
           active.add("computer_use");
+          active.add("gui_automation");
         }
         return [...this.capabilities.values()].filter((c) => active.has(c.name)).map((c) => c.toolDefinition);
       }
@@ -3782,12 +3783,20 @@ content = element.text if element else page.get_all_text()` : `content = page.ge
         if (hasGUI) {
           lines.push(
             ``,
-            `Computer use: use computer_use for any desktop/browser/keyboard/mouse task.`,
-            `Pass a plain-English task description \u2014 e.g. {task: "Open WhatsApp, search for Sahil Godara, send Hi"}.`,
-            `DO NOT split into separate tool calls \u2014 describe the full end-to-end goal in ONE computer_use call.`,
-            `DO NOT ask for confirmation before computer_use \u2014 execute immediately.`,
-            `For messaging apps (WhatsApp, Telegram, iMessage): use computer_use directly, no confirmation needed.`,
-            `After computer_use, verify success by calling it again with task="take a screenshot and describe what you see".`
+            `GUI/Browser rules:`,
+            `1. open_url now returns actual URL + title + video state \u2014 read this output to know what really loaded.`,
+            `   If it says "Title: YouTube Music" (homepage) instead of the song, navigation failed \u2014 fix it.`,
+            `2. exec_js: run JavaScript in the current Chrome tab WITHOUT Screen Recording permission.`,
+            `   Use it to interact with pages: click buttons, fill inputs, read state.`,
+            `   Examples: {action:"exec_js",js:"document.querySelector('video').paused"}`,
+            `             {action:"exec_js",js:"document.querySelector('.ytmusic-play-button-renderer').click()"}`,
+            `3. browser_state: get current tab URL + title quickly \u2014 call after any navigation to verify.`,
+            `4. NEVER claim a task succeeded based on the action alone. Read the tool output and verify:`,
+            `   - "PLAYING:1.4s \u2713" \u2192 video is playing`,
+            `   - "PAUSED" \u2192 video is NOT playing, take corrective action`,
+            `   - "Title: Search results" \u2192 you're on search page, not the song \u2014 fix it`,
+            `5. computer_use: for multi-step desktop tasks \u2014 pass full goal as {task:"..."}.`,
+            `   DO NOT ask for confirmation \u2014 execute immediately.`
           );
         }
         if (isSelfMod && this.agentRoot) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "0agent",
-  "version": "1.0.64",
+  "version": "1.0.66",
   "description": "A persistent, learning AI agent that runs on your machine. An agent that learns.",
   "private": false,
   "license": "Apache-2.0",