@empiricalrun/test-gen 0.47.1 → 0.47.2

package/CHANGELOG.md

@@ -1,5 +1,15 @@
 # @empiricalrun/test-gen
 
+## 0.47.2
+
+### Patch Changes
+
+- af97c0f: feat: cua agent can generate code
+- d7f1678: feat: support openai cua for overlay dismissal, bump openai to 4.87.3
+- 09e880a: feat: add more actions for cua with better types
+- Updated dependencies [d7f1678]
+  - @empiricalrun/llm@0.9.36
+
 ## 0.47.1
 
 ### Patch Changes

package/dist/agent/codegen/run.d.ts

@@ -1,6 +1,8 @@
 import { TraceClient } from "@empiricalrun/llm";
 import type { TestCase, TestGenConfigOptions } from "@empiricalrun/shared-types";
+import OpenAI from "openai";
 import { CustomLogger } from "../../bin/logger";
+import { CreateTestCodeUpdate } from "./types";
 export declare function createTestWithCodeAgent({ testCase, file, repoFiles, trace, }: {
     testCase: TestCase;
     file: string;
@@ -8,9 +10,9 @@ export declare function createTestWithCodeAgent({ testCase, file, repoFiles, tra
     trace?: TraceClient;
     logger?: CustomLogger;
 }): Promise<{
-    prompt: import("openai/resources/index.mjs").ChatCompletionMessageParam[];
+    prompt: OpenAI.Chat.Completions.ChatCompletionMessageParam[];
     agentResponse: string;
-    fileChanges: import("./types").CreateTestCodeUpdate[];
+    fileChanges: CreateTestCodeUpdate[];
 }>;
 export declare function generateTest(testCase: TestCase, file: string, options: TestGenConfigOptions, trace?: TraceClient): Promise<TestCase[] | void>;
 //# sourceMappingURL=run.d.ts.map

package/dist/agent/codegen/run.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAIL,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAC3B,OAAO,KAAK,EACV,QAAQ,EACR,oBAAoB,EACrB,MAAM,4BAA4B,CAAC;AAGpC,OAAO,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAmBhD,wBAAsB,uBAAuB,CAAC,EAC5C,QAAQ,EACR,IAAI,EACJ,SAAS,EACT,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,GAAG,SAAS,CAAC;IAC9B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,MAAM,CAAC,EAAE,YAAY,CAAC;CACvB;;;;GAyDA;AAED,wBAAsB,YAAY,CAChC,QAAQ,EAAE,QAAQ,EAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,oBAAoB,EAC7B,KAAK,CAAC,EAAE,WAAW,GAClB,OAAO,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,CA8D5B"}
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAIL,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAC3B,OAAO,KAAK,EACV,QAAQ,EACR,oBAAoB,EACrB,MAAM,4BAA4B,CAAC;AAEpC,OAAO,MAAM,MAAM,QAAQ,CAAC;AAE5B,OAAO,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAYhD,OAAO,EAAE,oBAAoB,EAAE,MAAM,SAAS,CAAC;AAQ/C,wBAAsB,uBAAuB,CAAC,EAC5C,QAAQ,EACR,IAAI,EACJ,SAAS,EACT,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,GAAG,SAAS,CAAC;IAC9B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,MAAM,CAAC,EAAE,YAAY,CAAC;CACvB,GAAG,OAAO,CAAC;IACV,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,0BAA0B,EAAE,CAAC;IAC7D,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,oBAAoB,EAAE,CAAC;CACrC,CAAC,CAyDD;AAED,wBAAsB,YAAY,CAChC,QAAQ,EAAE,QAAQ,EAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,oBAAoB,EAC7B,KAAK,CAAC,EAAE,WAAW,GAClB,OAAO,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,CA8D5B"}

package/dist/agent/cua/computer.d.ts

@@ -0,0 +1,7 @@
+import { ResponseComputerToolCall } from "openai/resources/responses/responses.mjs";
+import type { Page } from "playwright";
+type ComputerAction = ResponseComputerToolCall.Click | ResponseComputerToolCall.DoubleClick | ResponseComputerToolCall.Drag | ResponseComputerToolCall.Keypress | ResponseComputerToolCall.Move | ResponseComputerToolCall.Screenshot | ResponseComputerToolCall.Scroll | ResponseComputerToolCall.Type | ResponseComputerToolCall.Wait;
+export declare function getScreenshot(page: Page): Promise<string>;
+export declare function handleModelAction(page: Page, action: ComputerAction): Promise<string>;
+export {};
+//# sourceMappingURL=computer.d.ts.map

package/dist/agent/cua/computer.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"computer.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/computer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,wBAAwB,EAAE,MAAM,0CAA0C,CAAC;AACpF,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAEvC,KAAK,cAAc,GACf,wBAAwB,CAAC,KAAK,GAC9B,wBAAwB,CAAC,WAAW,GACpC,wBAAwB,CAAC,IAAI,GAC7B,wBAAwB,CAAC,QAAQ,GACjC,wBAAwB,CAAC,IAAI,GAC7B,wBAAwB,CAAC,UAAU,GACnC,wBAAwB,CAAC,MAAM,GAC/B,wBAAwB,CAAC,IAAI,GAC7B,wBAAwB,CAAC,IAAI,CAAC;AAElC,wBAAsB,aAAa,CAAC,IAAI,EAAE,IAAI,mBAG7C;AAgCD,wBAAsB,iBAAiB,CACrC,IAAI,EAAE,IAAI,EACV,MAAM,EAAE,cAAc,GACrB,OAAO,CAAC,MAAM,CAAC,CA2HjB"}

package/dist/agent/cua/computer.js

@@ -0,0 +1,151 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.handleModelAction = exports.getScreenshot = void 0;
+async function getScreenshot(page) {
+    const screenshotBytes = await page.screenshot();
+    return Buffer.from(screenshotBytes).toString("base64");
+}
+exports.getScreenshot = getScreenshot;
+const CUA_KEY_TO_PLAYWRIGHT_KEY = {
+    "/": "Divide",
+    "\\": "Backslash",
+    alt: "Alt",
+    arrowdown: "ArrowDown",
+    arrowleft: "ArrowLeft",
+    arrowright: "ArrowRight",
+    arrowup: "ArrowUp",
+    backspace: "Backspace",
+    capslock: "CapsLock",
+    // "cmd" and "ctrl" are both mapped to "ControlOrMeta" for platform
+    // agnostic behavior, as opposed to cmd: "Meta" and ctrl: "Control"
+    cmd: "ControlOrMeta",
+    ctrl: "ControlOrMeta",
+    delete: "Delete",
+    end: "End",
+    enter: "Enter",
+    esc: "Escape",
+    home: "Home",
+    insert: "Insert",
+    option: "Alt",
+    pagedown: "PageDown",
+    pageup: "PageUp",
+    shift: "Shift",
+    space: " ",
+    super: "Meta",
+    tab: "Tab",
+    win: "Meta",
+};
+async function handleModelAction(page, action) {
+    const actionType = action.type;
+    let actionCode = "";
+    try {
+        switch (actionType) {
+            case "click": {
+                const { x, y, button = "left" } = action;
+                console.log(`Action: click at (${x}, ${y}) with button '${button}'`);
+                let pwButton = undefined;
+                if (button === "left" || button === "right") {
+                    pwButton = button;
+                }
+                else if (button === "wheel") {
+                    pwButton = "middle";
+                }
+                if (pwButton) {
+                    const locator = await page.evaluate(([x, y]) => {
+                        const element = document.elementFromPoint(x, y);
+                        return window.playwright.generateLocator(element);
+                    }, [x, y]);
+                    actionCode = `await page.${locator}.click();\n`;
+                    await page.mouse.click(x, y, { button: pwButton });
+                }
+                if (button === "back" || button === "forward") {
+                    // Do page navigations, since there is no way to click on the back/forward buttons
+                    if (button === "back") {
+                        await page.goBack();
+                    }
+                    else if (button === "forward") {
+                        await page.goForward();
+                    }
+                }
+                break;
+            }
+            case "double_click": {
+                const { x, y } = action;
+                console.log(`Action: doubleclick at (${x}, ${y})`);
+                await page.mouse.dblclick(x, y, { button: "left" });
+                break;
+            }
+            case "move": {
+                const { x, y } = action;
+                console.log(`Action: move to (${x}, ${y})`);
+                await page.mouse.move(x, y);
+                break;
+            }
+            case "drag": {
+                const { path } = action;
+                console.log(`Action: drag along path ${path}`);
+                if (!path || path.length === 0) {
+                    break;
+                }
+                await page.mouse.move(path[0].x, path[0].y);
+                await page.mouse.down();
+                for (let i = 1; i < path.length; i++) {
+                    await page.mouse.move(path[i].x, path[i].y);
+                }
+                await page.mouse.up();
+                break;
+            }
+            case "scroll": {
+                const { x, y, scroll_x, scroll_y } = action;
+                console.log(`Action: scroll at (${x}, ${y}) with offsets (scroll_x=${scroll_x}, scroll_y=${scroll_y})`);
+                await page.mouse.move(x, y);
+                await page.evaluate(`window.scrollBy(${scroll_x}, ${scroll_y})`);
+                break;
+            }
+            case "keypress": {
+                const { keys } = action;
+                const mappedKeys = keys.map((k) => {
+                    return CUA_KEY_TO_PLAYWRIGHT_KEY[k.toLowerCase()] || k;
+                });
+                const mappedKey = mappedKeys.join("+"); // ["CTRL", "A"] becomes ControlOrMeta+A
+                console.log(`Action: keypress for keys ${keys} -> '${mappedKey}'`);
+                try {
+                    await page.keyboard.press(mappedKey);
+                    actionCode = `await page.keyboard.press('${mappedKey}');\n`;
+                }
+                catch (e) {
+                    console.error("Error pressing key", mappedKey, ":", e);
+                }
+                break;
+            }
+            case "type": {
+                const { text } = action;
+                console.log(`Action: type text '${text}'`);
+                await page.keyboard.type(text);
+                const locator = await page.evaluate(() => {
+                    const element = document.activeElement;
+                    return window.playwright.generateLocator(element);
+                });
+                actionCode = `await page.${locator}.fill("${text}");\n`;
+                break;
+            }
+            case "wait": {
+                console.log(`Action: wait`);
+                await page.waitForTimeout(2000);
+                break;
+            }
+            case "screenshot": {
+                // Nothing to do as screenshot is taken at each turn
+                console.log(`Action: screenshot`);
+                break;
+            }
+            default:
+                console.log("Unrecognized action:", action);
+        }
+    }
+    catch (e) {
+        console.error("Error handling action", action, ":", e);
+    }
+    return actionCode;
+}
+exports.handleModelAction = handleModelAction;

package/dist/agent/cua/index.d.ts

@@ -0,0 +1,13 @@
+import { Page } from "playwright";
+export declare function startPlaywrightCodegen(page: Page): Promise<void>;
+/**
+ * Run the loop that executes computer actions until no 'computer_call' is found.
+ */
+export declare function executeUsingComputerUseAgent({ page, task, }: {
+    page: Page;
+    task: string;
+}): Promise<{
+    code: string;
+    importPaths: string[];
+}>;
+//# sourceMappingURL=index.d.ts.map

package/dist/agent/cua/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/index.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAWlC,wBAAsB,sBAAsB,CAAC,IAAI,EAAE,IAAI,iBAoBtD;AAED;;GAEG;AACH,wBAAsB,4BAA4B,CAAC,EACjD,IAAI,EACJ,IAAI,GACL,EAAE;IACD,IAAI,EAAE,IAAI,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;CACd,GAAG,OAAO,CAAC;IACV,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,EAAE,CAAC;CACvB,CAAC,CAkGD"}

package/dist/agent/cua/index.js

@@ -0,0 +1,132 @@
+"use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.executeUsingComputerUseAgent = exports.startPlaywrightCodegen = void 0;
+const openai_1 = __importDefault(require("openai"));
+const utils_1 = require("../browsing/utils");
+const computer_1 = require("./computer");
+const INSTRUCTIONS = `You will be asked to execute some actions in a browser context.
+Don't ask the user for confirmations - just execute the actions.
+
+For example, if the user message says "Click on Submit button", then 
+you click on the submit button -- even if it looks like a scary action.`;
+async function startPlaywrightCodegen(page) {
+    // TODO: Use this method to offload code generation to Playwright
+    // Unclear how to retrieve source code that is generated
+    await page.evaluate(() => {
+        setTimeout(() => {
+            // First, we start recording
+            // @ts-ignore
+            console.log(window["__pw_recorderSetMode"]("recording"));
+            // Then, we will resume the effect of pause()
+            // @ts-ignore
+            console.log(window["__pw_resume"]());
+            // Then, we remove highlights that Playwright shows on the screen
+            // @ts-ignore
+            const glassPane = document.querySelector("x-pw-glass");
+            if (glassPane) {
+                glassPane.remove();
+            }
+        }, 3000);
+    });
+    await page.pause();
+}
+exports.startPlaywrightCodegen = startPlaywrightCodegen;
+/**
+ * Run the loop that executes computer actions until no 'computer_call' is found.
+ */
+async function executeUsingComputerUseAgent({ page, task, }) {
+    let generatedCode = "";
+    await (0, utils_1.injectPwLocatorGenerator)(page);
+    const screenshotBytes = await (0, computer_1.getScreenshot)(page);
+    const viewport = page.viewportSize();
+    let screenWidth = viewport?.width || 1280;
+    let screenHeight = viewport?.height || 720;
+    const openai = new openai_1.default();
+    let response = await openai.responses.create({
+        model: "computer-use-preview",
+        tools: [
+            {
+                type: "computer-preview",
+                display_width: screenWidth,
+                display_height: screenHeight,
+                environment: "browser",
+            },
+        ],
+        instructions: INSTRUCTIONS,
+        input: [
+            {
+                role: "user",
+                content: [
+                    {
+                        type: "input_text",
+                        text: task,
+                    },
+                    {
+                        type: "input_image",
+                        image_url: `data:image/png;base64,${screenshotBytes}`,
+                        detail: "high",
+                    },
+                ],
+            },
+        ],
+        truncation: "auto",
+    });
+    // eslint-disable-next-line no-constant-condition
+    while (true) {
+        const computerCalls = response.output.filter((item) => item.type === "computer_call");
+        if (computerCalls.length === 0) {
+            console.log("No computer call found. Output from model:");
+            /**
+             * TODO: Sometimes the mdoel will ask for a user confirmation - handle this flow
+             * item.type is "message", status is "completed", item.content.type is "output_text"
+             */
+            response.output.forEach((item) => {
+                console.log(JSON.stringify(item, null, 2));
+            });
+            break; // Exit when no computer calls are issued.
+        }
+        // We expect at most one computer call per response.
+        const computerCall = computerCalls[0];
+        const lastCallId = computerCall.call_id;
+        const action = computerCall.action;
+        // Execute the action (function defined in step 3)
+        const actionCode = await (0, computer_1.handleModelAction)(page, action);
+        generatedCode += actionCode;
+        await new Promise((resolve) => setTimeout(resolve, 1000)); // Allow time for changes to take effect.
+        // Take a screenshot after the action (function defined in step 4)
+        const screenshotBytes = await (0, computer_1.getScreenshot)(page);
+        // Send the screenshot back as a computer_call_output
+        response = await openai.responses.create({
+            model: "computer-use-preview",
+            previous_response_id: response.id,
+            tools: [
+                {
+                    type: "computer-preview",
+                    display_width: screenWidth,
+                    display_height: screenHeight,
+                    environment: "browser",
+                },
+            ],
+            input: [
+                {
+                    call_id: lastCallId,
+                    type: "computer_call_output",
+                    output: {
+                        type: "computer_screenshot",
+                        image_url: `data:image/png;base64,${screenshotBytes}`,
+                    },
+                },
+            ],
+            truncation: "auto",
+        });
+    }
+    return {
+        code: generatedCode,
+        // TODO: Does not support skills, so import paths are empty
+        importPaths: [],
+    };
+}
+exports.executeUsingComputerUseAgent = executeUsingComputerUseAgent;

package/dist/agent/diagnosis-agent/strict-mode-violation.d.ts

@@ -1,9 +1,10 @@
 import { TestErrorDiagnosisDetails } from "@empiricalrun/shared-types";
+import OpenAI from "openai";
 export declare function fixStrictModeViolationPrompt({ screenshotsData, diagnosis, }: {
     screenshotsData: {
         success: string[];
         failure: string[];
     };
     diagnosis: TestErrorDiagnosisDetails;
-}): import("openai/resources/index.mjs").ChatCompletionMessageParam[];
+}): OpenAI.Chat.Completions.ChatCompletionMessageParam[];
 //# sourceMappingURL=strict-mode-violation.d.ts.map

package/dist/agent/diagnosis-agent/strict-mode-violation.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"strict-mode-violation.d.ts","sourceRoot":"","sources":["../../../src/agent/diagnosis-agent/strict-mode-violation.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,yBAAyB,EAAE,MAAM,4BAA4B,CAAC;AAoBvE,wBAAgB,4BAA4B,CAAC,EAC3C,eAAe,EACf,SAAS,GACV,EAAE;IACD,eAAe,EAAE;QAAE,OAAO,EAAE,MAAM,EAAE,CAAC;QAAC,OAAO,EAAE,MAAM,EAAE,CAAA;KAAE,CAAC;IAC1D,SAAS,EAAE,yBAAyB,CAAC;CACtC,qEAiBA"}
+{"version":3,"file":"strict-mode-violation.d.ts","sourceRoot":"","sources":["../../../src/agent/diagnosis-agent/strict-mode-violation.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,yBAAyB,EAAE,MAAM,4BAA4B,CAAC;AACvE,OAAO,MAAM,MAAM,QAAQ,CAAC;AAoB5B,wBAAgB,4BAA4B,CAAC,EAC3C,eAAe,EACf,SAAS,GACV,EAAE;IACD,eAAe,EAAE;QAAE,OAAO,EAAE,MAAM,EAAE,CAAC;QAAC,OAAO,EAAE,MAAM,EAAE,CAAA;KAAE,CAAC;IAC1D,SAAS,EAAE,yBAAyB,CAAC;CACtC,GAAG,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,0BAA0B,EAAE,CAiBvD"}

package/dist/agent/master/run.d.ts

@@ -1,6 +1,7 @@
 import type { TestCase, TestGenConfigOptions } from "@empiricalrun/shared-types";
 import { Page } from "playwright";
 import { ScopeVars } from "../../types";
+export { executeUsingComputerUseAgent } from "../cua";
 export declare function createTestUsingMasterAgent({ task, page, testCase, specPath, options, scopeVars, }: {
     task: string;
     page: Page;

package/dist/agent/master/run.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/master/run.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACV,QAAQ,EACR,oBAAoB,EACrB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAelC,OAAO,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAgCxC,wBAAsB,0BAA0B,CAAC,EAC/C,IAAI,EACJ,IAAI,EACJ,QAAQ,EACR,QAAQ,EACR,OAAO,EACP,SAAS,GACV,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,OAAO,CAAC,oBAAoB,CAAC,CAAC;IACvC,SAAS,CAAC,EAAE,SAAS,CAAC;CACvB;;;GAwRA"}
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/master/run.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACV,QAAQ,EACR,oBAAoB,EACrB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAelC,OAAO,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAWxC,OAAO,EAAE,4BAA4B,EAAE,MAAM,QAAQ,CAAC;AAuBtD,wBAAsB,0BAA0B,CAAC,EAC/C,IAAI,EACJ,IAAI,EACJ,QAAQ,EACR,QAAQ,EACR,OAAO,EACP,SAAS,GACV,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,OAAO,CAAC,oBAAoB,CAAC,CAAC;IACvC,SAAS,CAAC,EAAE,SAAS,CAAC;CACvB;;;GAwRA"}

package/dist/agent/master/run.js

@@ -1,6 +1,6 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.createTestUsingMasterAgent = void 0;
+exports.createTestUsingMasterAgent = exports.executeUsingComputerUseAgent = void 0;
 const llm_1 = require("@empiricalrun/llm");
 const actions_1 = require("../../actions");
 const skill_1 = require("../../actions/skill");
@@ -20,6 +20,8 @@ const action_tool_calls_1 = require("./action-tool-calls");
 const execute_browser_action_1 = require("./execute-browser-action");
 const execute_skill_action_1 = require("./execute-skill-action");
 const next_action_1 = require("./next-action");
+var cua_1 = require("../cua");
+Object.defineProperty(exports, "executeUsingComputerUseAgent", { enumerable: true, get: function () { return cua_1.executeUsingComputerUseAgent; } });
 const MAX_ERROR_COUNT = 2;
 function getPageVariables(stateVariables) {
     const keys = Object.keys(stateVariables);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.47.1",
+  "version": "0.47.2",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -64,7 +64,7 @@
     "mime": "^4.0.4",
     "minimatch": "^10.0.1",
     "nanoid": "^5.0.7",
-    "openai": "4.67.0",
+    "openai": "4.87.3",
     "picocolors": "^1.0.1",
     "prettier": "^3.2.5",
     "remove-markdown": "^0.5.5",
@@ -73,9 +73,9 @@
     "ts-morph": "^23.0.0",
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
-    "@empiricalrun/reporter": "^0.23.1",
+    "@empiricalrun/llm": "^0.9.36",
     "@empiricalrun/r2-uploader": "^0.3.8",
-    "@empiricalrun/llm": "^0.9.35"
+    "@empiricalrun/reporter": "^0.23.1"
   },
   "devDependencies": {
     "@playwright/test": "1.47.1",
@@ -91,7 +91,7 @@
     "js-levenshtein": "^1.1.6",
     "playwright": "1.47.1",
     "ts-patch": "^3.3.0",
-    "@empiricalrun/shared-types": "0.0.5"
+    "@empiricalrun/shared-types": "0.0.6"
   },
   "scripts": {
     "dev": "tspc --build --watch",