@browserbasehq/stagehand 3.0.7-alpha-ddd7464487a0cd0615d4f1ba458bc6f3fa51b27d → 3.0.7-alpha-bd2db925f66a826d61d58be1611d55646cbdb560

package/dist/index.js

@@ -179,7 +179,7 @@ var __forAwait = (obj, it, method) => (it = obj[__knownSymbol("asyncIterator")])
 var STAGEHAND_VERSION;
 var init_version = __esm({
   "lib/version.ts"() {
-    STAGEHAND_VERSION = "3.0.7-alpha-ddd7464487a0cd0615d4f1ba458bc6f3fa51b27d";
+    STAGEHAND_VERSION = "3.0.7-alpha-bd2db925f66a826d61d58be1611d55646cbdb560";
   }
 });
 
@@ -27179,7 +27179,6 @@ init_sdkErrors();
 var import_genai = require("@google/genai");
 var import_zod2 = require("zod");
 var import_v3 = __toESM(require("zod/v3"));
-var import_sharp = __toESM(require("sharp"));
 var ID_PATTERN = /^\d+-\d+$/;
 var zFactories = {
   v4: import_zod2.z,
@@ -33509,6 +33508,7 @@ var OpenAICUAClient = class extends AgentClient {
   constructor(type, modelName, userProvidedInstructions, clientOptions, tools) {
     super(type, modelName, userProvidedInstructions);
     this.currentViewport = { width: 1288, height: 711 };
+    this.actualScreenshotSize = { width: 1288, height: 711 };
     this.reasoningItems = /* @__PURE__ */ new Map();
     this.environment = "browser";
     this.apiKey = (clientOptions == null ? void 0 : clientOptions.apiKey) || process.env.OPENAI_API_KEY || "";
@@ -33529,6 +33529,9 @@ var OpenAICUAClient = class extends AgentClient {
   setViewport(width, height) {
     this.currentViewport = { width, height };
   }
+  setScreenshotSize(width, height) {
+    this.actualScreenshotSize = { width, height };
+  }
   setCurrentUrl(url) {
     this.currentUrl = url;
   }
@@ -33963,9 +33966,16 @@ var OpenAICUAClient = class extends AgentClient {
   }
   convertComputerCallToAction(call) {
     const { action } = call;
+    const scaledAction = __spreadValues({}, action);
+    if (action.x !== void 0 && action.y !== void 0) {
+      const scaleX = this.currentViewport.width / this.actualScreenshotSize.width;
+      const scaleY = this.currentViewport.height / this.actualScreenshotSize.height;
+      scaledAction.x = Math.floor(action.x * scaleX);
+      scaledAction.y = Math.floor(action.y * scaleY);
+    }
     return __spreadValues({
       type: action.type
-    }, action);
+    }, scaledAction);
   }
   convertFunctionCallToAction(call) {
     try {
@@ -34184,6 +34194,7 @@ var GoogleCUAClient = class extends AgentClient {
   constructor(type, modelName, userProvidedInstructions, clientOptions, tools) {
     super(type, modelName, userProvidedInstructions);
     this.currentViewport = { width: 1288, height: 711 };
+    this.actualScreenshotSize = { width: 1288, height: 711 };
     this.history = [];
     this.environment = "ENVIRONMENT_BROWSER";
     this.tools = tools;
@@ -34222,6 +34233,9 @@ var GoogleCUAClient = class extends AgentClient {
   setViewport(width, height) {
     this.currentViewport = { width, height };
   }
+  setScreenshotSize(width, height) {
+    this.actualScreenshotSize = { width, height };
+  }
   setCurrentUrl(url) {
     this.currentUrl = url;
   }
@@ -34837,9 +34851,13 @@ var GoogleCUAClient = class extends AgentClient {
   normalizeCoordinates(x2, y) {
     x2 = Math.min(999, Math.max(0, x2));
     y = Math.min(999, Math.max(0, y));
+    const screenshotX = x2 / 1e3 * this.actualScreenshotSize.width;
+    const screenshotY = y / 1e3 * this.actualScreenshotSize.height;
+    const scaleX = this.currentViewport.width / this.actualScreenshotSize.width;
+    const scaleY = this.currentViewport.height / this.actualScreenshotSize.height;
     return {
-      x: Math.floor(x2 / 1e3 * this.currentViewport.width),
-      y: Math.floor(y / 1e3 * this.currentViewport.height)
+      x: Math.floor(screenshotX * scaleX),
+      y: Math.floor(screenshotY * scaleY)
     };
   }
   captureScreenshot(options) {
@@ -34890,6 +34908,8 @@ var MicrosoftCUAClient = class extends AgentClient {
     };
     // Resized dimensions for model input
     this.resizedViewport = { width: 1288, height: 711 };
+    // Actual screenshot dimensions (tracked separately from viewport)
+    this.actualScreenshotSize = { width: 1288, height: 711 };
     this.apiKey = (clientOptions == null ? void 0 : clientOptions.apiKey) || process.env.AZURE_API_KEY || process.env.FIREWORKS_API_KEY || "";
     this.baseURL = (clientOptions == null ? void 0 : clientOptions.baseURL) || process.env.AZURE_ENDPOINT || process.env.FIREWORKS_ENDPOINT || "";
     this.clientOptions = {
@@ -34916,6 +34936,9 @@ var MicrosoftCUAClient = class extends AgentClient {
     this.currentViewport = { width, height };
     this.resizedViewport = this.smartResize(width, height);
   }
+  setScreenshotSize(width, height) {
+    this.actualScreenshotSize = { width, height };
+  }
   setCurrentUrl(url) {
     this.currentUrl = url;
   }
@@ -34953,7 +34976,7 @@ var MicrosoftCUAClient = class extends AgentClient {
    * Simplified to match Python's minimal approach
    */
   generateSystemPrompt() {
-    const { width, height } = this.resizedViewport;
+    const { width, height } = this.actualScreenshotSize;
     let basePrompt = "You are a helpful assistant.";
     if (this.userProvidedInstructions) {
       basePrompt = `${basePrompt}
@@ -35114,8 +35137,8 @@ ${functionCallTemplate}`;
     const transformCoordinate = (coord) => {
       if (!coord || coord.length !== 2) return coord;
       const [x2, y] = coord;
-      const scaleX = this.currentViewport.width / this.resizedViewport.width;
-      const scaleY = this.currentViewport.height / this.resizedViewport.height;
+      const scaleX = this.currentViewport.width / this.actualScreenshotSize.width;
+      const scaleY = this.currentViewport.height / this.actualScreenshotSize.height;
       return [Math.round(x2 * scaleX), Math.round(y * scaleY)];
     };
     const baseAction = {
@@ -35624,6 +35647,14 @@ var AgentProvider = class _AgentProvider {
 // lib/v3/handlers/v3CuaAgentHandler.ts
 init_flowLogger();
 init_sdkErrors();
+function getPNGDimensions(buffer) {
+  if (buffer.length < 24 || buffer[0] !== 137 || buffer[1] !== 80 || buffer[2] !== 78 || buffer[3] !== 71) {
+    throw new Error("Invalid PNG file");
+  }
+  const width = buffer.readUInt32BE(16);
+  const height = buffer.readUInt32BE(20);
+  return { width, height };
+}
 var V3CuaAgentHandler = class {
   constructor(v3, logger, options, tools) {
     this.v3 = v3;
@@ -35653,8 +35684,23 @@ var V3CuaAgentHandler = class {
     this.agentClient.setScreenshotProvider(() => __async(this, null, function* () {
       this.ensureNotClosed();
       const page = yield this.v3.context.awaitActivePage();
-      const base64 = yield page.screenshot({ fullPage: false });
-      return base64.toString("base64");
+      const screenshotBuffer = yield page.screenshot({ fullPage: false });
+      if (this.agentClient instanceof GoogleCUAClient || this.agentClient instanceof OpenAICUAClient || this.agentClient instanceof MicrosoftCUAClient) {
+        try {
+          const dimensions = getPNGDimensions(screenshotBuffer);
+          this.agentClient.setScreenshotSize(
+            dimensions.width,
+            dimensions.height
+          );
+        } catch (e2) {
+          this.logger({
+            category: "agent",
+            message: `Could not read screenshot dimensions: ${e2}`,
+            level: 1
+          });
+        }
+      }
+      return screenshotBuffer.toString("base64");
     }));
     this.agentClient.setActionHandler((action) => __async(this, null, function* () {
       var _a4, _b, _c;
@@ -36078,11 +36124,26 @@ var V3CuaAgentHandler = class {
       });
       try {
         const page = yield this.v3.context.awaitActivePage();
-        const base64Image = yield page.screenshot({ fullPage: false });
-        this.v3.bus.emit("agent_screensot_taken_event", base64Image);
+        const screenshotBuffer = yield page.screenshot({ fullPage: false });
+        if (this.agentClient instanceof GoogleCUAClient || this.agentClient instanceof OpenAICUAClient || this.agentClient instanceof MicrosoftCUAClient) {
+          try {
+            const dimensions = getPNGDimensions(screenshotBuffer);
+            this.agentClient.setScreenshotSize(
+              dimensions.width,
+              dimensions.height
+            );
+          } catch (e2) {
+            this.logger({
+              category: "agent",
+              message: `Could not read screenshot dimensions: ${e2}`,
+              level: 1
+            });
+          }
+        }
+        this.v3.bus.emit("agent_screenshot_taken_event", screenshotBuffer);
         const currentUrl = page.url();
         return yield this.agentClient.captureScreenshot({
-          base64Image,
+          base64Image: screenshotBuffer.toString("base64"),
           currentUrl
         });
       } catch (e2) {
@@ -66748,7 +66809,7 @@ var _V3 = class _V3 {
     }
     this.logger({
       category: "agent",
-      message: `Creating v3 agent instance with options: ${JSON.stringify(options)}`,
+      message: "Creating v3 agent instance",
       level: 1,
       auxiliary: __spreadValues({
         cua: { value: isCuaMode ? "true" : "false", type: "boolean" },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@browserbasehq/stagehand",
-  "version": "3.0.7-alpha-ddd7464487a0cd0615d4f1ba458bc6f3fa51b27d",
+  "version": "3.0.7-alpha-bd2db925f66a826d61d58be1611d55646cbdb560",
   "description": "An AI web browsing framework focused on simplicity and extensibility.",
   "main": "./dist/index.js",
   "module": "./dist/index.js",
@@ -41,7 +41,6 @@
     "pino": "^9.6.0",
     "pino-pretty": "^13.0.0",
     "playwright": "^1.52.0",
-    "sharp": "^0.34.5",
     "ws": "^8.18.0",
     "zod-to-json-schema": "^3.25.0"
   },