strix-agent 0.1.1__py3-none-any.whl

strix/tools/browser/browser_actions_schema.xml

@@ -0,0 +1,183 @@
+<?xml version="1.0" ?>
+<tools>
+  <tool name="browser_action">
+    <description>Perform browser actions using a Playwright-controlled browser with multiple tabs.
+  The browser is PERSISTENT and remains active until explicitly closed, allowing for
+  multi-step workflows and long-running processes across multiple tabs.</description>
+    <parameters>
+      <parameter name="action" type="string" required="true">
+      </parameter>
+      <parameter name="url" type="string" required="false">
+        <description>Required for 'launch', 'goto', and optionally for 'new_tab' actions. The URL to launch the browser at, navigate to, or load in new tab. Must include appropriate protocol (e.g., http://, https://, file://).</description>
+      </parameter>
+      <parameter name="coordinate" type="string" required="false">
+        <description>Required for 'click', 'double_click', and 'hover' actions. Format: "x,y" (e.g., "432,321"). Coordinates should target the center of elements (buttons, links, etc.). Must be within the browser viewport resolution. Be very careful to calculate the coordinates correctly based on the previous screenshot.</description>
+      </parameter>
+      <parameter name="text" type="string" required="false">
+        <description>Required for 'type' action. The text to type in the field.</description>
+      </parameter>
+      <parameter name="tab_id" type="string" required="false">
+        <description>Required for 'switch_tab' and 'close_tab' actions. Optional for other actions to specify which tab to operate on. The ID of the tab to operate on. The first tab created during 'launch' has ID "tab_1". If not provided, actions will operate on the currently active tab.</description>
+      </parameter>
+      <parameter name="js_code" type="string" required="false">
+        <description>Required for 'execute_js' action. JavaScript code to execute in the page context. The code runs in the context of the current page and has access to the DOM and all page-defined variables and functions. The last evaluated expression's value is returned in the response.</description>
+      </parameter>
+      <parameter name="duration" type="string" required="false">
+        <description>Required for 'wait' action. Number of seconds to pause execution. Can be fractional (e.g., 0.5 for half a second).</description>
+      </parameter>
+      <parameter name="key" type="string" required="false">
+        <description>Required for 'press_key' action. The key to press. Valid values include: - Single characters: 'a'-'z', 'A'-'Z', '0'-'9' - Special keys: 'Enter', 'Escape', 'ArrowLeft', 'ArrowRight', etc. - Modifier keys: 'Shift', 'Control', 'Alt', 'Meta' - Function keys: 'F1'-'F12'</description>
+      </parameter>
+      <parameter name="file_path" type="string" required="false">
+        <description>Required for 'save_pdf' action. The file path where to save the PDF.</description>
+      </parameter>
+      <parameter name="clear" type="boolean" required="false">
+        <description>For 'get_console_logs' action: whether to clear console logs after retrieving them. Default is False (keep logs).</description>
+      </parameter>
+    </parameters>
+    <returns type="Dict[str, Any]">
+      <description>Response containing: - screenshot: Base64 encoded PNG of the current page state - url: Current page URL - title: Current page title - viewport: Current browser viewport dimensions - tab_id: ID of the current active tab - all_tabs: Dict of all open tab IDs and their URLs - message: Status message about the action performed - js_result: Result of JavaScript execution (for execute_js action) - pdf_saved: File path of saved PDF (for save_pdf action) - console_logs: Array of console messages (for get_console_logs action)   Limited to 50KB total and 200 most recent logs. Individual messages truncated at 1KB. - page_source: HTML source code (for view_source action)   Large pages are truncated to 100KB (keeping beginning and end sections).</description>
+    </returns>
+    <notes>
+  Important usage rules:
+  1. PERSISTENCE: The browser remains active and maintains its state until
+     explicitly closed with the 'close' action. This allows for multi-step workflows
+     across multiple tool calls and tabs.
+  2. Browser interaction MUST start with 'launch' and end with 'close'.
+  3. Only one action can be performed per call.
+  4. To visit a new URL not reachable from current page, either:
+     - Use 'goto' action
+     - Open a new tab with the URL
+     - Close browser and relaunch
+  5. Click coordinates must be derived from the most recent screenshot.
+  6. You MUST click on the center of the element, not the edge. You MUST calculate
+     the coordinates correctly based on the previous screenshot, otherwise the click
+     will fail. After clicking, check the new screenshot to verify the click was
+     successful.
+  7. Tab management:
+     - First tab from 'launch' is "tab_1"
+     - New tabs are numbered sequentially ("tab_2", "tab_3", etc.)
+     - Must have at least one tab open at all times
+     - Actions affect the currently active tab unless tab_id is specified
+  8. JavaScript execution (following Playwright evaluation patterns):
+     - Code runs in the browser page context, not the tool context
+     - Has access to DOM (document, window, etc.) and page variables/functions
+     - The LAST EVALUATED EXPRESSION is automatically returned - no return statement needed
+     - For simple values: document.title (returns the title)
+     - For objects: {title: document.title, url: location.href} (returns the object)
+     - For async operations: Use await and the promise result will be returned
+     - AVOID explicit return statements - they can break evaluation
+     - object literals must be wrapped in paranthesis when they are the final expression
+     - Variables from tool context are NOT available - pass data as parameters if needed
+     - Examples of correct patterns:
+       * Single value: document.querySelectorAll('img').length
+       * Object result: {images: document.images.length, links: document.links.length}
+       * Async operation: await fetch(location.href).then(r => r.status)
+       * DOM manipulation: document.body.style.backgroundColor = 'red'; 'background changed'
+
+  9. Wait action:
+     - Time is specified in seconds
+     - Can be used to wait for page loads, animations, etc.
+     - Can be fractional (e.g., 0.5 seconds)
+     - Screenshot is captured after the wait
+  10. The browser can operate concurrently with other tools. You may invoke
+      terminal, python, or other tools (in separate assistant messages) while maintaining
+      the active browser session, enabling sophisticated multi-tool workflows.
+  11. Keyboard actions:
+      - Use press_key for individual key presses
+      - Use type for typing regular text
+      - Some keys have special names based on Playwright's key documentation
+  12. All code in the js_code parameter is executed as-is - there's no need to
+      escape special characters or worry about formatting. Just write your JavaScript
+      code normally. It can be single line or multi-line.
+  13. For form filling, click on the field first, then use 'type' to enter text.
+  14. The browser runs in headless mode using Chrome engine for security and performance.
+    </notes>
+    <examples>
+  # Launch browser at URL (creates tab_1)
+  <function=browser_action>
+  <parameter=action>launch</parameter>
+  <parameter=url>https://example.com</parameter>
+  </function>
+
+  # Navigate to different URL
+  <function=browser_action>
+  <parameter=action>goto</parameter>
+  <parameter=url>https://github.com</parameter>
+  </function>
+
+  # Open new tab with different URL
+  <function=browser_action>
+  <parameter=action>new_tab</parameter>
+  <parameter=url>https://another-site.com</parameter>
+  </function>
+
+  # Wait for page load
+  <function=browser_action>
+  <parameter=action>wait</parameter>
+  <parameter=duration>2.5</parameter>
+  </function>
+
+  # Click login button at coordinates from screenshot
+  <function=browser_action>
+  <parameter=action>click</parameter>
+  <parameter=coordinate>450,300</parameter>
+  </function>
+
+  # Click username field and type
+  <function=browser_action>
+  <parameter=action>click</parameter>
+  <parameter=coordinate>400,200</parameter>
+  </function>
+
+  <function=browser_action>
+  <parameter=action>type</parameter>
+  <parameter=text>user@example.com</parameter>
+  </function>
+
+  # Click password field and type
+  <function=browser_action>
+  <parameter=action>click</parameter>
+  <parameter=coordinate>400,250</parameter>
+  </function>
+
+  <function=browser_action>
+  <parameter=action>type</parameter>
+  <parameter=text>mypassword123</parameter>
+  </function>
+
+  # Press Enter key
+  <function=browser_action>
+  <parameter=action>press_key</parameter>
+  <parameter=key>Enter</parameter>
+  </function>
+
+  # Execute JavaScript to get page stats (correct pattern - no return statement)
+  <function=browser_action>
+  <parameter=action>execute_js</parameter>
+  <parameter=js_code>const images = document.querySelectorAll('img');
+const links = document.querySelectorAll('a');
+{
+    images: images.length,
+    links: links.length,
+    title: document.title
+}</parameter>
+  </function>
+
+  # Scroll down
+  <function=browser_action>
+  <parameter=action>scroll_down</parameter>
+  </function>
+
+  # Get console logs
+  <function=browser_action>
+  <parameter=action>get_console_logs</parameter>
+  </function>
+
+  # View page source
+  <function=browser_action>
+  <parameter=action>view_source</parameter>
+  </function>
+    </examples>
+  </tool>
+</tools>