@browserbasehq/orca 3.5.1-preview.0 → 3.7.0-preview.1

package/dist/cjs/lib/v3/agent/GoogleCUAClient.js

@@ -174,6 +174,8 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
         await this.initializeHistory(instruction);
         let totalInputTokens = 0;
         let totalOutputTokens = 0;
+        let totalReasoningTokens = 0;
+        let totalCachedInputTokens = 0;
         let totalInferenceTime = 0;
         try {
             // Execute steps until completion or max steps reached
@@ -187,6 +189,8 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                 const result = await this.executeStep(logger);
                 totalInputTokens += result.usage.input_tokens;
                 totalOutputTokens += result.usage.output_tokens;
+                totalReasoningTokens += result.usage.reasoning_tokens;
+                totalCachedInputTokens += result.usage.cached_input_tokens;
                 totalInferenceTime += result.usage.inference_time_ms;
                 // Add actions to the list
                 actions.push(...result.actions);
@@ -209,6 +213,8 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                 usage: {
                     input_tokens: totalInputTokens,
                     output_tokens: totalOutputTokens,
+                    reasoning_tokens: totalReasoningTokens,
+                    cached_input_tokens: totalCachedInputTokens,
                     inference_time_ms: totalInferenceTime,
                 },
             };
@@ -228,6 +234,8 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                 usage: {
                     input_tokens: totalInputTokens,
                     output_tokens: totalOutputTokens,
+                    reasoning_tokens: totalReasoningTokens,
+                    cached_input_tokens: totalCachedInputTokens,
                     inference_time_ms: totalInferenceTime,
                 },
             };
@@ -365,9 +373,14 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
             }
             // Execute actions and collect function responses
             const functionResponses = [];
-            if (result.actions.length > 0) {
+            // Always process the model's response, even when it produced no
+            // executable action (e.g. a standalone take_screenshot). The model
+            // expects exactly one function response — a fresh screenshot — per
+            // computer-use call; skipping it when there are 0 actions leaves the
+            // model blind and it gives up after one empty turn.
+            {
                 let hasError = false;
-                // Execute all actions
+                // Execute all actions (a no-op when there are none).
                 for (let i = 0; i < result.actions.length; i++) {
                     const action = result.actions[i];
                     logger({
@@ -385,6 +398,15 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                             level: 2,
                         });
                     }
+                    else if (action.type === "screenshot") {
+                        // No interaction to perform — the screenshot captured below is
+                        // returned to the model as this call's function response.
+                        logger({
+                            category: "agent",
+                            message: "take_screenshot: capturing current page",
+                            level: 2,
+                        });
+                    }
                     else if (action.type === "custom_tool") {
                         const toolName = action.name;
                         const toolArgs = action.arguments;
@@ -438,7 +460,7 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                                 level: 2,
                             });
                             const screenshot = await this.captureScreenshot();
-                            const base64Data = screenshot.replace(/^data:image\/png;base64,/, "");
+                            const base64Data = screenshot.base64;
                             // Create one function response for each computer use function call
                             // Following Python SDK pattern: FunctionResponse with parts containing inline_data
                             for (const functionCall of computerUseFunctionCalls) {
@@ -460,7 +482,7 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                                         parts: [
                                             {
                                                 inlineData: {
-                                                    mimeType: "image/png",
+                                                    mimeType: screenshot.mediaType,
                                                     data: base64Data,
                                                 },
                                             },
@@ -497,8 +519,14 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                 message: result.message,
                 completed: result.completed,
                 usage: {
+                    // promptTokenCount is the TOTAL input and already includes the
+                    // cached portion; cachedContentTokenCount is the cache-hit subset
+                    // (tracked separately for visibility, not additive). thoughtsTokenCount
+                    // is Gemini's thinking/reasoning output.
                     input_tokens: usageMetadata?.promptTokenCount || 0,
                     output_tokens: usageMetadata?.candidatesTokenCount || 0,
+                    reasoning_tokens: usageMetadata?.thoughtsTokenCount || 0,
+                    cached_input_tokens: usageMetadata?.cachedContentTokenCount || 0,
                     inference_time_ms: elapsedMs,
                 },
             };
@@ -553,23 +581,26 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                     level: 2,
                 });
                 // Convert function call to action(s)
-                const action = this.convertFunctionCallToAction(part.functionCall);
+                const action = this.convertFunctionCallToAction(part.functionCall, logger);
                 if (action) {
-                    // Special handling for type_text_at - we need to click first
-                    if (part.functionCall.name === "type_text_at" &&
-                        action.type === "type") {
+                    // Special handling for type actions. gemini-2.5 type_text_at carries
+                    // coordinates (click the field first); gemini-3.x `type` has none and
+                    // types into the element the model already focused.
+                    if (action.type === "type") {
                         logger({
                             category: "agent",
                             message: `Adding action: ${JSON.stringify(action)}`,
                             level: 2,
                         });
-                        // First add a click action at the same coordinates
-                        actions.push({
-                            type: "click",
-                            x: action.x,
-                            y: action.y,
-                            button: "left",
-                        });
+                        // Click the target first only when coordinates are provided.
+                        if (typeof action.x === "number" && typeof action.y === "number") {
+                            actions.push({
+                                type: "click",
+                                x: action.x,
+                                y: action.y,
+                                button: "left",
+                            });
+                        }
                         // If clear_before_typing is true (default), add a select all
                         if (action.clearBeforeTyping) {
                             // Select all text in the field
@@ -623,18 +654,80 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
     /**
      * Convert Google function call to Stagehand action
      */
-    convertFunctionCallToAction(functionCall) {
-        const { name, args } = functionCall;
-        if (!name || !args) {
+    convertFunctionCallToAction(functionCall, logger) {
+        const { name: rawName } = functionCall;
+        // Default args to an empty object so no-argument predefined functions
+        // (e.g. take_screenshot, go_back) are not rejected by the guard below.
+        const args = functionCall.args ?? {};
+        if (!rawName) {
             return null;
         }
+        // The gemini-3.x computer-use tool renamed several of the predefined
+        // functions that gemini-2.5-computer-use-preview used, and adds a
+        // descriptive `intent` arg to every call. The argument fields are otherwise
+        // unchanged (confirmed for click/navigate). Normalize the 3.x names to the
+        // canonical 2.5 names so a single set of handlers serves both generations;
+        // `intent` is ignored since handlers only read the args they need.
+        // NOTE: click and take_screenshot are confirmed from live gemini-3.5-flash
+        // traffic; the rest are inferred from the same drop-the-qualifier pattern
+        // and are safe aliases (any unmapped name still hits the warning below).
+        // gemini-3.x renamed several predefined functions vs gemini-2.5. Alias the
+        // 3.x names whose behavior maps cleanly onto a 2.5 canonical handler. Click
+        // variants that carry distinct semantics (double/triple/right/middle click,
+        // move) are NOT aliased here — they have dedicated cases below so the click
+        // count and button are preserved. 2.5 never emits any of these names, so
+        // the 2.5 handlers are unaffected.
+        const NAME_ALIASES = {
+            click: "click_at",
+            left_click: "click_at",
+            type: "type_text_at",
+            type_text: "type_text_at",
+            hover: "hover_at",
+            scroll: "scroll_at",
+            drag: "drag_and_drop",
+            key: "key_combination",
+            keys: "key_combination",
+            key_press: "key_combination",
+            press_key: "key_combination",
+            press_keys: "key_combination",
+            hotkey: "key_combination",
+            screenshot: "take_screenshot",
+            wait: "wait_5_seconds",
+        };
+        const name = NAME_ALIASES[rawName] ?? rawName;
+        // Predefined computer-use tools take precedence over custom tools. If a
+        // custom tool was registered under a reserved (aliased) name, note that the
+        // predefined tool wins rather than silently dropping the custom one.
+        if (rawName in NAME_ALIASES && (0, googleCustomToolHandler_js_1.isCustomTool)(functionCall, this.tools)) {
+            logger?.({
+                category: "agent",
+                message: `Custom tool "${rawName}" collides with a predefined Google CUA function; using the predefined tool. Rename the custom tool to avoid the conflict.`,
+                level: 2,
+            });
+        }
         switch (name) {
             case "open_web_browser":
                 return {
                     type: "open_web_browser",
                     timestamp: Date.now(),
                 };
+            case "take_screenshot":
+                // No UI interaction. The step loop captures a fresh screenshot and
+                // returns it to the model as this call's function response; marked as a
+                // recognized action (not null) so it isn't dropped, and skipped in the
+                // executor like open_web_browser.
+                return {
+                    type: "screenshot",
+                    timestamp: Date.now(),
+                };
             case "click_at": {
+                // x/y are required; reject a malformed call rather than normalizing
+                // undefined/NaN/Infinity into bad coordinates. (gemini-3.x `click`
+                // aliases here.)
+                if (!GoogleCUAClient.isFiniteCoord(args.x) ||
+                    !GoogleCUAClient.isFiniteCoord(args.y)) {
+                    return null;
+                }
                 const { x, y } = this.normalizeCoordinates(args.x, args.y);
                 return {
                     type: "click",
@@ -643,28 +736,74 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                     button: args.button || "left",
                 };
             }
-            case "type_text_at": {
+            // gemini-3.x click family. The executor natively supports double/triple
+            // click and right/middle button, so preserve those semantics rather than
+            // collapsing to a single left click. All require integer coordinates;
+            // drop the call (return null) on a malformed payload so the executor is
+            // never handed NaN.
+            case "double_click":
+            case "triple_click":
+            case "right_click":
+            case "middle_click":
+            case "move": {
+                if (!GoogleCUAClient.isFiniteCoord(args.x) ||
+                    !GoogleCUAClient.isFiniteCoord(args.y)) {
+                    return null;
+                }
                 const { x, y } = this.normalizeCoordinates(args.x, args.y);
-                // Google's type_text_at includes press_enter and clear_before_typing parameters
+                if (name === "move") {
+                    return { type: "move", x, y };
+                }
+                if (name === "double_click" || name === "triple_click") {
+                    return { type: name, x, y };
+                }
+                return {
+                    type: "click",
+                    x,
+                    y,
+                    button: name === "right_click" ? "right" : "middle",
+                };
+            }
+            case "type_text_at": {
+                // text is required; reject a malformed call rather than typing
+                // "undefined". An empty string is valid (e.g. clear the field).
+                if (typeof args.text !== "string") {
+                    return null;
+                }
+                // press_enter and clear_before_typing are shared across generations.
                 const pressEnter = args.press_enter ?? false;
                 const clearBeforeTyping = args.clear_before_typing ?? true;
-                // For type_text_at, we need to click first then type
-                // This matches the behavior expected by Google's CUA
-                // We'll handle this in the executeStep method by converting to two actions
-                return {
+                const base = {
                     type: "type",
                     text: args.text,
-                    x,
-                    y,
                     pressEnter,
                     clearBeforeTyping,
                 };
+                // gemini-2.5 type_text_at carries x/y (click the field, then type).
+                // gemini-3.x `type` has no coordinates and types into the element the
+                // model already focused with a preceding click. Only attach coords when
+                // present so the executor doesn't click at NaN.
+                if (typeof args.x === "number" && typeof args.y === "number") {
+                    const { x, y } = this.normalizeCoordinates(args.x, args.y);
+                    return { ...base, x, y };
+                }
+                return base;
             }
             case "key_combination": {
-                const keys = args.keys
-                    .split("+")
-                    .map((key) => key.trim())
+                // gemini-2.5 key_combination sends `keys` as a "+"-joined string;
+                // gemini-3.x `hotkey` sends a `keys` array, and `press_key` sends a
+                // single `key`. Accept all three.
+                const raw = args.keys !== undefined ? args.keys : args.key;
+                const parts = Array.isArray(raw)
+                    ? raw
+                    : String(raw ?? "").split("+");
+                const keys = parts
+                    .map((key) => String(key).trim())
+                    .filter(Boolean)
                     .map((key) => (0, cuaKeyMapping_js_1.mapKeyToPlaywright)(key));
+                if (keys.length === 0) {
+                    return null;
+                }
                 return {
                     type: "keypress",
                     keys,
@@ -678,9 +817,24 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                 };
             }
             case "scroll_at": {
+                // A coordinate-less (or malformed) `scroll` alias (gemini-3.x) scrolls
+                // the document via PageUp/PageDown.
+                if (!GoogleCUAClient.isFiniteCoord(args.x) ||
+                    !GoogleCUAClient.isFiniteCoord(args.y)) {
+                    const dir = (args.direction || "down").toLowerCase();
+                    return {
+                        type: "keypress",
+                        keys: [dir === "up" ? "PageUp" : "PageDown"],
+                    };
+                }
                 const { x, y } = this.normalizeCoordinates(args.x, args.y);
                 const direction = (args.direction || "down").toLowerCase();
-                const magnitude = typeof args.magnitude === "number" ? args.magnitude : 800;
+                // 2.5 uses `magnitude`; gemini-3.x `scroll` uses `magnitude_in_pixels`.
+                const magnitude = GoogleCUAClient.isFiniteCoord(args.magnitude)
+                    ? args.magnitude
+                    : GoogleCUAClient.isFiniteCoord(args.magnitude_in_pixels)
+                        ? args.magnitude_in_pixels
+                        : 800;
                 let scroll_x = 0;
                 let scroll_y = 0;
                 if (direction === "up") {
@@ -708,6 +862,11 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                 };
             }
             case "navigate":
+                // url is required; reject a malformed call rather than navigating to
+                // "undefined".
+                if (typeof args.url !== "string" || args.url.length === 0) {
+                    return null;
+                }
                 return {
                     type: "goto",
                     url: args.url,
@@ -739,8 +898,20 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                     url: "https://www.google.com",
                 };
             case "drag_and_drop": {
-                const startPoint = this.normalizeCoordinates(args.x, args.y);
-                const endPoint = this.normalizeCoordinates(args.destination_x, args.destination_y);
+                // 2.5 uses x/y + destination_x/destination_y; gemini-3.x `drag_and_drop`
+                // uses start_x/start_y + end_x/end_y. Accept either.
+                const sx = args.x ?? args.start_x;
+                const sy = args.y ?? args.start_y;
+                const ex = args.destination_x ?? args.end_x;
+                const ey = args.destination_y ?? args.end_y;
+                if (!GoogleCUAClient.isFiniteCoord(sx) ||
+                    !GoogleCUAClient.isFiniteCoord(sy) ||
+                    !GoogleCUAClient.isFiniteCoord(ex) ||
+                    !GoogleCUAClient.isFiniteCoord(ey)) {
+                    return null;
+                }
+                const startPoint = this.normalizeCoordinates(sx, sy);
+                const endPoint = this.normalizeCoordinates(ex, ey);
                 return {
                     type: "drag",
                     path: [
@@ -763,6 +934,14 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
                 return null;
         }
     }
+    /**
+     * True only for a usable coordinate/number: rejects undefined, non-numbers,
+     * and the numeric edge cases NaN and Infinity (both `typeof "number"`), so
+     * malformed function calls are dropped instead of normalizing into NaN.
+     */
+    static isFiniteCoord(value) {
+        return typeof value === "number" && Number.isFinite(value);
+    }
     /**
      * Normalize coordinates from Google's 0-1000 range to viewport dimensions
      */
@@ -781,13 +960,15 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
         }
         // Use provided options if available
         if (options?.base64Image) {
-            return `data:image/png;base64,${options.base64Image}`;
+            return {
+                base64: options.base64Image,
+                mediaType: options.mediaType ?? "image/png",
+            };
         }
         // Use the screenshot provider if available
         if (this.screenshotProvider) {
             try {
-                const base64Image = await this.screenshotProvider();
-                return `data:image/png;base64,${base64Image}`;
+                return await this.screenshotProvider();
             }
             catch (error) {
                 console.error("Error capturing screenshot:", error);
@@ -795,7 +976,7 @@ class GoogleCUAClient extends AgentClient_js_1.AgentClient {
             }
         }
         throw new sdkErrors_js_1.AgentScreenshotProviderError("`screenshotProvider` has not been set. " +
-            "Please call `setScreenshotProvider()` with a valid function that returns a base64-encoded image");
+            "Please call `setScreenshotProvider()` with a valid function that returns a base64-encoded image and media type");
     }
 }
 exports.GoogleCUAClient = GoogleCUAClient;