illuma-agents 1.0.20 → 1.0.21

package/src/tools/BrowserTools.ts

@@ -476,24 +476,29 @@ export function createBrowserGetPageStateTool(): DynamicStructuredTool<typeof Br
         requiresBrowserExecution: true,
         // Special flag: extension should inject fresh context into the conversation
         requiresContextRefresh: true,
+        // IMPORTANT: Tell the agent to wait
+        message: 'Page state is being captured by the browser extension. The element list will be provided in the next message. DO NOT proceed with click or type actions until you receive the actual element list.',
       });
     },
     {
       name: EBrowserTools.GET_PAGE_STATE,
       description: `Get fresh page state showing current interactive elements.
 
-**CRITICAL**: You MUST call this tool after:
-- browser_navigate (to see elements on the new page)
-- browser_click (if it caused navigation or page changes)
-- Any action that might have changed the visible elements
+**CRITICAL WORKFLOW**: After calling this tool, you MUST STOP and WAIT. The browser extension will capture the page state and return the element list. DO NOT plan any browser_click or browser_type actions in the same response - you don't have the element indices yet!
 
-This tool returns the updated list of interactive elements with their [index] numbers.
-Without calling this after navigation, you will NOT know what elements exist on the new page.
+**When to use**:
+- After browser_navigate (to see elements on the new page)
+- After browser_click (if it caused navigation or page changes)
+- Any time you need to see what elements are currently on the page
 
-**Workflow example**:
-1. browser_navigate to amazon.com
-2. browser_get_page_state to see Amazon's elements
-3. Now you can see the search input's [index] and use browser_type
+**IMPORTANT**: This tool captures the page state asynchronously. The actual element list will be provided AFTER this tool completes. You should:
+1. Call this tool
+2. STOP and wait for the response with the element list
+3. In your NEXT response, use the element indices for click/type actions
+
+Example workflow:
+- Turn 1: browser_navigate to amazon.com, then browser_get_page_state
+- Turn 2: (After receiving element list) browser_type with the correct search input index
 
 Example: browser_get_page_state({ reason: "to see elements after navigation" })`,
       schema: BrowserGetPageStateSchema,

package/src/types/run.ts

@@ -115,6 +115,14 @@ export type RunConfig = {
   returnContent?: boolean;
   tokenCounter?: TokenCounter;
   indexTokenCountMap?: Record<string, number>;
+  /** 
+   * Enable browser extension mode with interrupt-based tool execution.
+   * When true:
+   * - Uses MemorySaver checkpointer for pause/resume
+   * - Browser tools will interrupt execution and wait for extension results
+   * - Extension must call resume endpoint with Command to continue
+   */
+  browserMode?: boolean;
 };
 
 export type ProvidedCallbacks =