@empiricalrun/test-gen 0.64.3 → 0.66.0

package/dist/agent/cua/pw-codegen/pw-pause/index.js

@@ -3,45 +3,67 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.PlaywrightPauseCodegen = exports.PW_PAUSE_IPC_PORT = exports.revertToOriginalPwCode = void 0;
-exports.preparePlaywrightForCodegen = preparePlaywrightForCodegen;
+exports.PlaywrightPauseCodegen = exports.revertToOriginalPwCode = exports.preparePlaywrightForCodegen = void 0;
+exports.canUsePauseCodegen = canUsePauseCodegen;
 const express_1 = __importDefault(require("express"));
-const fs_1 = __importDefault(require("fs"));
-const path_1 = __importDefault(require("path"));
-const utils_1 = require("./utils");
-var utils_2 = require("./utils");
-Object.defineProperty(exports, "revertToOriginalPwCode", { enumerable: true, get: function () { return utils_2.revertToOriginalPwCode; } });
-exports.PW_PAUSE_IPC_PORT = 3039;
-async function preparePlaywrightForCodegen(repoDir) {
-    const npmListOutput = await (0, utils_1.runNpmList)(repoDir);
-    if (!npmListOutput) {
-        return;
+const ipc_1 = require("./ipc");
+const patch_1 = require("./patch");
+var patch_2 = require("./patch");
+Object.defineProperty(exports, "preparePlaywrightForCodegen", { enumerable: true, get: function () { return patch_2.preparePlaywrightForCodegen; } });
+Object.defineProperty(exports, "revertToOriginalPwCode", { enumerable: true, get: function () { return patch_2.revertToOriginalPwCode; } });
+async function canUsePauseCodegen(repoDir) {
+    const hasPatchedPwCode = await (0, patch_1.hasPatchedRecorderApp)(repoDir);
+    if (!hasPatchedPwCode) {
+        // This codegen approach requires patching the playwright code to setup an
+        // IPC server <> client that passes generated code from Playwright to @empiricalrun/test-gen
+        console.log("Cannot use pause codegen because recorder app is not patched");
+        return false;
     }
-    const playwrightCorePath = (0, utils_1.getPlaywrightCoreFromNpmList)(npmListOutput);
-    const pathToRecorderApp = path_1.default.join(playwrightCorePath, "lib", "server", "recorder", "recorderApp.js");
-    if (!fs_1.default.existsSync(pathToRecorderApp)) {
-        const errMsg = `Cannot patch Playwright: ${pathToRecorderApp} does not exist`;
-        throw new Error(errMsg);
+    if (!(await (0, ipc_1.isIPCPortAvailable)())) {
+        // The IPC port is hardcoded, and it should be available for IPC to work
+        console.log("Cannot use pause codegen because IPC port is not available");
+        return false;
     }
-    await (0, utils_1.createFileBackup)(pathToRecorderApp);
-    const port = exports.PW_PAUSE_IPC_PORT;
-    await (0, utils_1.patchPwCode)(pathToRecorderApp, port);
+    const isRunningHeaded = process.env.RUN_PLAYWRIGHT_HEADED &&
+        process.env.RUN_PLAYWRIGHT_HEADED === "true";
+    if (!isRunningHeaded) {
+        // This codegen approach requires running Playwright in headed mode because
+        // it depends on the page.pause() API, which is no-op in headless mode
+        console.log("Cannot use pause codegen because RUN_PLAYWRIGHT_HEADED is not true");
+        return false;
+    }
+    // eslint-disable-next-line turbo/no-undeclared-env-vars
+    if (!process.env.PW_CODEGEN_NO_INSPECTOR) {
+        // PW_CODEGEN_NO_INSPECTOR should be set to true, so that Playwright runs the
+        // EmptyRecorderApp() -- which is the entrypoint to our patched code
+        // Ref: https://github.com/microsoft/playwright/blob/c96558d0691c0377dbc0c772a1bd8120d546e6d7/packages/playwright-core/src/server/recorder/recorderApp.ts#L98
+        console.log("Cannot use pause codegen because PW_CODEGEN_NO_INSPECTOR is not set");
+        return false;
+    }
+    // Need all conditions to be met to use pause codegen
+    return true;
 }
-// TODO: Fix this to implement BasePlaywrightCodegen
 class PlaywrightPauseCodegen {
     port = 0;
     page;
     server;
+    codeForLastAction;
     constructor() {
-        this.port = exports.PW_PAUSE_IPC_PORT;
+        this.port = ipc_1.PW_PAUSE_IPC_PORT;
+    }
+    async saveCode(code) {
+        const generatedCode = code.map((c) => c.actions.join("\n")).join("\n");
+        if (generatedCode) {
+            this.codeForLastAction = generatedCode;
+        }
     }
     async initialize(page) {
         // Start server to receive generated code from patch
         const app = (0, express_1.default)();
         app.use(express_1.default.json());
-        app.post("/test", async (req, res) => {
-            console.log("--- Received request in PlaywrightPatchCodegen ---");
-            console.log(req.body);
+        app.post("/sources", async (req, res) => {
+            const { payload } = req.body;
+            await this.saveCode(JSON.parse(payload));
             return res.send({ success: true });
         });
         await new Promise((resolve) => {
@@ -74,11 +96,17 @@ class PlaywrightPauseCodegen {
         await page.pause();
     }
     async recordAction() {
-        console.log("Record action is no-op");
+        // Record action is no-op
     }
-    async getCodegenResult() {
-        console.log("Get codegen result is no-op");
-        return "";
+    async getCodeForLastAction() {
+        // Wait for the server to send the code
+        await new Promise((resolve) => setTimeout(resolve, 1000));
+        let codeToReturn = this.codeForLastAction;
+        this.codeForLastAction = undefined;
+        if (!codeToReturn) {
+            throw new Error("No code for last action");
+        }
+        return codeToReturn;
     }
 }
 exports.PlaywrightPauseCodegen = PlaywrightPauseCodegen;

package/dist/agent/cua/pw-codegen/pw-pause/ipc.d.ts

@@ -0,0 +1,3 @@
+export declare const PW_PAUSE_IPC_PORT = 3039;
+export declare function isIPCPortAvailable(): Promise<boolean>;
+//# sourceMappingURL=ipc.d.ts.map

package/dist/agent/cua/pw-codegen/pw-pause/ipc.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ipc.d.ts","sourceRoot":"","sources":["../../../../../src/agent/cua/pw-codegen/pw-pause/ipc.ts"],"names":[],"mappings":"AAEA,eAAO,MAAM,iBAAiB,OAAO,CAAC;AAEtC,wBAAsB,kBAAkB,qBAGvC"}

package/dist/agent/cua/pw-codegen/pw-pause/ipc.js

@@ -0,0 +1,13 @@
+"use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.PW_PAUSE_IPC_PORT = void 0;
+exports.isIPCPortAvailable = isIPCPortAvailable;
+const detect_port_1 = __importDefault(require("detect-port"));
+exports.PW_PAUSE_IPC_PORT = 3039;
+async function isIPCPortAvailable() {
+    const portAvailable = await (0, detect_port_1.default)(exports.PW_PAUSE_IPC_PORT);
+    return portAvailable === exports.PW_PAUSE_IPC_PORT;
+}

package/dist/agent/cua/pw-codegen/pw-pause/{utils.d.ts → patch.d.ts}

@@ -1,16 +1,9 @@
-/**
- * Helpful commands for debugging:
- *
- * 1. Find playwright-core path
- * cd `npm ls playwright-core --json --long | jq -r '.dependencies["@playwright/test"].dependencies.playwright.dependencies["playwright-core"].path'`
- *
- * 2. Find the path to the recorder app
- * cat lib/server/recorder/recorderApp.js
- *
- */
+export declare function preparePlaywrightForCodegen(repoDir: string): Promise<void>;
+export declare function getPathToRecorderApp(repoDir: string): Promise<string | undefined>;
 export declare function runNpmList(repoDir: string): Promise<string | undefined>;
 export declare function getPlaywrightCoreFromNpmList(output: string): any;
+export declare function hasPatchedRecorderApp(repoDir: string): Promise<boolean>;
 export declare function revertToOriginalPwCode(repoDir: string): Promise<void>;
 export declare function createFileBackup(pathToFile: string): Promise<void>;
 export declare function patchPwCode(pathToRecorderApp: string, port: number): Promise<void>;
-//# sourceMappingURL=utils.d.ts.map
+//# sourceMappingURL=patch.d.ts.map

package/dist/agent/cua/pw-codegen/pw-pause/patch.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"patch.d.ts","sourceRoot":"","sources":["../../../../../src/agent/cua/pw-codegen/pw-pause/patch.ts"],"names":[],"mappings":"AAmBA,wBAAsB,2BAA2B,CAAC,OAAO,EAAE,MAAM,iBAgBhE;AAED,wBAAsB,oBAAoB,CAAC,OAAO,EAAE,MAAM,+BAczD;AAED,wBAAsB,UAAU,CAAC,OAAO,EAAE,MAAM,+BAiB/C;AAED,wBAAgB,4BAA4B,CAAC,MAAM,EAAE,MAAM,OAuB1D;AAED,wBAAsB,qBAAqB,CAAC,OAAO,EAAE,MAAM,oBAY1D;AAED,wBAAsB,sBAAsB,CAAC,OAAO,EAAE,MAAM,iBAoB3D;AAED,wBAAsB,gBAAgB,CAAC,UAAU,EAAE,MAAM,iBAGxD;AAED,wBAAsB,WAAW,CAAC,iBAAiB,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,iBAkBxE"}

package/dist/agent/cua/pw-codegen/pw-pause/{utils.js → patch.js}

@@ -3,14 +3,18 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.preparePlaywrightForCodegen = preparePlaywrightForCodegen;
+exports.getPathToRecorderApp = getPathToRecorderApp;
 exports.runNpmList = runNpmList;
 exports.getPlaywrightCoreFromNpmList = getPlaywrightCoreFromNpmList;
+exports.hasPatchedRecorderApp = hasPatchedRecorderApp;
 exports.revertToOriginalPwCode = revertToOriginalPwCode;
 exports.createFileBackup = createFileBackup;
 exports.patchPwCode = patchPwCode;
 const child_process_1 = require("child_process");
 const fs_1 = __importDefault(require("fs"));
 const path_1 = __importDefault(require("path"));
+const ipc_1 = require("./ipc");
 /**
  * Helpful commands for debugging:
  *
@@ -21,6 +25,34 @@ const path_1 = __importDefault(require("path"));
  * cat lib/server/recorder/recorderApp.js
  *
  */
+const CUSTOM_PATCH_COMMENT = `// Custom patch code set by @empiricalrun/test-gen`;
+async function preparePlaywrightForCodegen(repoDir) {
+    try {
+        const pathToRecorderApp = await getPathToRecorderApp(repoDir);
+        if (!pathToRecorderApp) {
+            throw new Error("Cannot find path to recorder app");
+        }
+        if (!fs_1.default.existsSync(pathToRecorderApp)) {
+            const errMsg = `Cannot patch Playwright: ${pathToRecorderApp} does not exist`;
+            throw new Error(errMsg);
+        }
+        await createFileBackup(pathToRecorderApp);
+        await patchPwCode(pathToRecorderApp, ipc_1.PW_PAUSE_IPC_PORT);
+    }
+    catch (error) {
+        console.error("Error patching Playwright", error);
+        throw error;
+    }
+}
+async function getPathToRecorderApp(repoDir) {
+    const npmListOutput = await runNpmList(repoDir);
+    if (!npmListOutput) {
+        return;
+    }
+    const playwrightCorePath = getPlaywrightCoreFromNpmList(npmListOutput);
+    const pathToRecorderApp = path_1.default.join(playwrightCorePath, "lib", "server", "recorder", "recorderApp.js");
+    return pathToRecorderApp;
+}
 async function runNpmList(repoDir) {
     try {
         const output = (0, child_process_1.execSync)("npm ls playwright-core --json --long --production=false", {
@@ -57,6 +89,20 @@ function getPlaywrightCoreFromNpmList(output) {
     }
     return playwrightCoreDep.path;
 }
+async function hasPatchedRecorderApp(repoDir) {
+    try {
+        const pathToRecorderApp = await getPathToRecorderApp(repoDir);
+        if (!pathToRecorderApp) {
+            return false;
+        }
+        const recorderAppContents = fs_1.default.readFileSync(pathToRecorderApp, "utf-8");
+        return recorderAppContents.includes(CUSTOM_PATCH_COMMENT);
+    }
+    catch (error) {
+        console.error("Error checking if recorder app is patched", error);
+        return false;
+    }
+}
 async function revertToOriginalPwCode(repoDir) {
     try {
         const npmListOutput = await runNpmList(repoDir);
@@ -80,11 +126,11 @@ async function createFileBackup(pathToFile) {
 async function patchPwCode(pathToRecorderApp, port) {
     const currentFileContents = fs_1.default.readFileSync(pathToRecorderApp, "utf-8");
     const original = `async setSources(sources) {}`;
-    // TODO: Can we have a request payload and parse it in our code
     const replacement = `async setSources(sources) {
-    const payload = JSON.stringify(sources.filter(source => source.isRecorded).filter(source => source.id === "javascript").map(source => source.actions));
+    ${CUSTOM_PATCH_COMMENT}
+    const payload = JSON.stringify(sources.filter(source => source.isRecorded).filter(source => source.id === "javascript"));
     const body = JSON.stringify({ payload });
-    await fetch("http://localhost:${port}/test", {
+    await fetch("http://localhost:${port}/sources", {
       method: "POST",
       headers: {
         "Content-Type": "application/json",

package/dist/bin/index.js

@@ -35,7 +35,7 @@ function setupProcessListeners(cleanup) {
         events.forEach((event) => process.removeListener(event, cleanup));
     };
 }
-async function runChatAgent({ modelInput, chatSessionId, useDiskForChatState, initialPromptPath, withRetry, }) {
+async function runChatAgent({ modelInput, chatSessionId, useDiskForChatState, initialPromptPath, }) {
     if (modelInput && !utils_2.ARGS_TO_MODEL_MAP[modelInput]) {
         throw new Error(`Invalid chat model: ${modelInput}`);
     }
@@ -63,7 +63,6 @@ async function runChatAgent({ modelInput, chatSessionId, useDiskForChatState, in
         selectedModel: specifiedModel || defaultModel,
         useDiskForChatState: useDiskForChatState || false,
         initialPromptContent,
-        withRetry,
     });
 }
 async function runAgentsWorkflow(testGenConfig, testGenToken) {
@@ -182,7 +181,7 @@ async function runAgentsWorkflow(testGenConfig, testGenToken) {
             repoDir: process.cwd(),
             testFilePath: specPath,
         });
-        const { isError, error } = await (0, run_1.generateTestsUsingMasterAgent)({
+        const { isError, error } = await (0, run_1.runBrowsingAgent)({
             testCaseName: testCase.name,
             testCaseSuites: testCase.suites,
             testFilePath: specPath,
@@ -212,7 +211,7 @@ async function main() {
         .option("--use-chat", "Use chat agent (and not the workflow)")
         .option("--chat-session-id <chat-session-id>", "Identifier for chat session (fetched from dash.empirical.run)")
         .option("--use-disk-for-chat-state", "Save and load chat state from disk")
-        .option("--chat-model <model>", "Chat model to use (claude-3-7-sonnet-20250219 or claude-3-5-sonnet-20241022 or gemini-2.5-pro-preview-03-25)")
+        .option("--chat-model <model>", "Chat model to use (claude-3-7-sonnet-20250219 or claude-3-5-sonnet-20241022 or gemini-2.5-pro-preview-06-05)")
         .option("--initial-prompt <path>", "Path to an initial prompt file (e.g. prompt.md)")
         .option("--with-retry", "Use the retry strategy")
         .parse(process.argv);
@@ -249,7 +248,6 @@ async function main() {
             modelInput: completedOptions.chatModel,
             useDiskForChatState: completedOptions.useDiskForChatState,
             initialPromptPath: completedOptions.initialPrompt,
-            withRetry: completedOptions.withRetry,
         });
         return;
     }

package/dist/bin/utils/index.d.ts

@@ -1,4 +1,4 @@
-import { SupportedChatModels } from "@empiricalrun/llm/chat";
+import { SupportedChatModels } from "@empiricalrun/shared-types";
 export declare const ARGS_TO_MODEL_MAP: Record<string, SupportedChatModels>;
 export interface CLIOptions {
     token?: string;
@@ -11,7 +11,6 @@ export interface CLIOptions {
     initialPrompt?: string;
     chatSessionId?: string;
     chatModel?: (typeof ARGS_TO_MODEL_MAP)[keyof typeof ARGS_TO_MODEL_MAP];
-    withRetry?: boolean;
 }
 export declare function validateAndCompleteCliOptions(options: CLIOptions): Promise<CLIOptions>;
 export declare function printBanner(): void;

package/dist/bin/utils/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/bin/utils/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AAG7D,eAAO,MAAM,iBAAiB,EAAE,MAAM,CAAC,MAAM,EAAE,mBAAmB,CAajE,CAAC;AAEF,MAAM,WAAW,UAAU;IACzB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,SAAS,CAAC,EAAE,CAAC,OAAO,iBAAiB,CAAC,CAAC,MAAM,OAAO,iBAAiB,CAAC,CAAC;IACvE,SAAS,CAAC,EAAE,OAAO,CAAC;CACrB;AAQD,wBAAsB,6BAA6B,CACjD,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC,UAAU,CAAC,CAyDrB;AAED,wBAAgB,WAAW,SAgC1B"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/bin/utils/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,4BAA4B,CAAC;AAGjE,eAAO,MAAM,iBAAiB,EAAE,MAAM,CAAC,MAAM,EAAE,mBAAmB,CAgBjE,CAAC;AAEF,MAAM,WAAW,UAAU;IACzB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,SAAS,CAAC,EAAE,CAAC,OAAO,iBAAiB,CAAC,CAAC,MAAM,OAAO,iBAAiB,CAAC,CAAC;CACxE;AAQD,wBAAsB,6BAA6B,CACjD,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC,UAAU,CAAC,CAyDrB;AAED,wBAAgB,WAAW,SAgC1B"}

package/dist/bin/utils/index.js

@@ -13,9 +13,11 @@ exports.ARGS_TO_MODEL_MAP = {
     "claude-4": "claude-sonnet-4-20250514",
     "claude-sonnet-4": "claude-sonnet-4-20250514",
     "claude-opus-4": "claude-opus-4-20250514",
-    "gemini-2.5": "gemini-2.5-pro-preview-03-25",
-    "gemini-2.5-pro": "gemini-2.5-pro-preview-03-25",
-    "gemini-2.5-pro-preview-03-25": "gemini-2.5-pro-preview-03-25",
+    "gemini-2.5": "gemini-2.5-pro-preview-06-05",
+    "gemini-2.5-pro": "gemini-2.5-pro-preview-06-05",
+    "gemini-2.5-pro-preview-03-25": "gemini-2.5-pro-preview-06-05",
+    o3: "o3-2025-04-16",
+    "o3-2025-04-16": "o3-2025-04-16",
     "o4-mini": "o4-mini-2025-04-16",
     "o4-mini-2025-04-16": "o4-mini-2025-04-16",
 };

package/dist/file/client.d.ts

@@ -1,17 +1,11 @@
-import { ArtifactInput } from "@empiricalrun/shared-types";
+import { BrowserAgentIPCPayload } from "./server";
 declare class FileServiceClient {
     baseUrl: string;
     port: number | undefined;
     constructor();
     static isAvailable(): boolean;
-    updateTest({ generatedCode, task, importPaths, actionsSummary, }: {
-        generatedCode: string;
-        task: string;
-        importPaths: string[];
-        actionsSummary?: string;
-    }): Promise<any>;
+    sendAgentResult(payload: BrowserAgentIPCPayload): Promise<any>;
     post(path: string, body: any): Promise<any>;
-    sendArtifactInputsToServer(artifacts: ArtifactInput[]): Promise<any>;
 }
 export default FileServiceClient;
 //# sourceMappingURL=client.d.ts.map

package/dist/file/client.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../../src/file/client.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAE3D,cAAM,iBAAiB;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,GAAG,SAAS,CAAC;;IAUzB,MAAM,CAAC,WAAW;IAIZ,UAAU,CAAC,EACf,aAAa,EACb,IAAI,EACJ,WAAW,EACX,cAAc,GACf,EAAE;QACD,aAAa,EAAE,MAAM,CAAC;QACtB,IAAI,EAAE,MAAM,CAAC;QACb,WAAW,EAAE,MAAM,EAAE,CAAC;QACtB,cAAc,CAAC,EAAE,MAAM,CAAC;KACzB;IASK,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG;IAiB5B,0BAA0B,CAAC,SAAS,EAAE,aAAa,EAAE;CAgB5D;AAED,eAAe,iBAAiB,CAAC"}
+{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../../src/file/client.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,sBAAsB,EAAE,MAAM,UAAU,CAAC;AAElD,cAAM,iBAAiB;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,GAAG,SAAS,CAAC;;IAUzB,MAAM,CAAC,WAAW;IAIZ,eAAe,CAAC,OAAO,EAAE,sBAAsB;IAI/C,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG;CAgBnC;AAED,eAAe,iBAAiB,CAAC"}

package/dist/file/client.js

@@ -13,13 +13,8 @@ class FileServiceClient {
     static isAvailable() {
         return !!Number(process.env.IPC_FILE_SERVICE_PORT);
     }
-    async updateTest({ generatedCode, task, importPaths, actionsSummary, }) {
-        return this.post("/test", {
-            generatedCode,
-            task,
-            importPaths,
-            actionsSummary,
-        });
+    async sendAgentResult(payload) {
+        return this.post("/agent-results", payload);
     }
     async post(path, body) {
         const resp = await fetch(`${this.baseUrl}${path}`, {
@@ -37,21 +32,5 @@ class FileServiceClient {
             return data;
         }
     }
-    async sendArtifactInputsToServer(artifacts) {
-        const resp = await fetch(`${this.baseUrl}/artifact`, {
-            method: "POST",
-            headers: {
-                "Content-Type": "application/json",
-            },
-            body: JSON.stringify(artifacts),
-        });
-        if (!resp.ok) {
-            throw new Error(`API failed with status ${resp.statusText}`);
-        }
-        else {
-            const data = await resp.json();
-            return data;
-        }
-    }
 }
 exports.default = FileServiceClient;

package/dist/file/server.d.ts

@@ -1,20 +1,32 @@
-import { ArtifactInput } from "@empiricalrun/shared-types";
+import { ArtifactInput, Usage } from "@empiricalrun/shared-types";
+import { BrowserAgentResult } from "../agent/cua";
+export type BrowserAgentIPCPayload = {
+    task: string;
+    result: BrowserAgentResult;
+    usage: Usage;
+    generatedCode: string;
+    importPaths: string[];
+};
 export declare class FileServiceServer {
     private port;
     private filePath;
     private repoDir;
     private server;
-    private actionsSummary;
     private updateFile;
     private onComplete?;
     private artifactsInputs;
+    private result;
+    private usage;
     constructor({ port, repoDir, updateFile, onComplete, }: {
         port: number;
         repoDir: string;
         updateFile: boolean;
         onComplete?: () => void;
     });
-    getActionsSummary(): string | undefined;
+    getResultAndUsage(): {
+        result: BrowserAgentResult | undefined;
+        usage: Usage | undefined;
+    };
     getArtifactInputsFromServer(): ArtifactInput[];
     setFilePath(filePath: string): void;
     startFileService(): Promise<number>;

package/dist/file/server.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/file/server.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAY3D,qBAAa,iBAAiB;IAC5B,OAAO,CAAC,IAAI,CAAa;IACzB,OAAO,CAAC,QAAQ,CAAc;IAC9B,OAAO,CAAC,OAAO,CAAc;IAC7B,OAAO,CAAC,MAAM,CAA4C;IAC1D,OAAO,CAAC,cAAc,CAAqB;IAC3C,OAAO,CAAC,UAAU,CAAkB;IACpC,OAAO,CAAC,UAAU,CAAC,CAAa;IAChC,OAAO,CAAC,eAAe,CAAuB;gBAClC,EACV,IAAI,EACJ,OAAO,EACP,UAAU,EACV,UAAU,GACX,EAAE;QACD,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,UAAU,EAAE,OAAO,CAAC;QACpB,UAAU,CAAC,EAAE,MAAM,IAAI,CAAC;KACzB;IAOD,iBAAiB;IAIjB,2BAA2B;IAI3B,WAAW,CAAC,QAAQ,EAAE,MAAM;IAItB,gBAAgB,IAAI,OAAO,CAAC,MAAM,CAAC;IA4EnC,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC;CAgB5B"}
+{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/file/server.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,KAAK,EAAE,MAAM,4BAA4B,CAAC;AAKlE,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAQlD,MAAM,MAAM,sBAAsB,GAAG;IACnC,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,kBAAkB,CAAC;IAC3B,KAAK,EAAE,KAAK,CAAC;IACb,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,MAAM,EAAE,CAAC;CACvB,CAAC;AAEF,qBAAa,iBAAiB;IAC5B,OAAO,CAAC,IAAI,CAAa;IACzB,OAAO,CAAC,QAAQ,CAAc;IAC9B,OAAO,CAAC,OAAO,CAAc;IAC7B,OAAO,CAAC,MAAM,CAA4C;IAC1D,OAAO,CAAC,UAAU,CAAkB;IACpC,OAAO,CAAC,UAAU,CAAC,CAAa;IAEhC,OAAO,CAAC,eAAe,CAAuB;IAC9C,OAAO,CAAC,MAAM,CAAiC;IAC/C,OAAO,CAAC,KAAK,CAAoB;gBAErB,EACV,IAAI,EACJ,OAAO,EACP,UAAU,EACV,UAAU,GACX,EAAE;QACD,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,UAAU,EAAE,OAAO,CAAC;QACpB,UAAU,CAAC,EAAE,MAAM,IAAI,CAAC;KACzB;IAOD,iBAAiB;;;;IAIjB,2BAA2B;IAI3B,WAAW,CAAC,QAAQ,EAAE,MAAM;IAItB,gBAAgB,IAAI,OAAO,CAAC,MAAM,CAAC;IA4DnC,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC;CAgB5B"}

package/dist/file/server.js

@@ -14,18 +14,19 @@ class FileServiceServer {
     filePath = "";
     repoDir = "";
     server;
-    actionsSummary;
     updateFile = false;
     onComplete;
     artifactsInputs = [];
+    result;
+    usage;
     constructor({ port, repoDir, updateFile, onComplete, }) {
         this.port = port;
         this.repoDir = repoDir;
         this.updateFile = updateFile;
         this.onComplete = onComplete;
     }
-    getActionsSummary() {
-        return this.actionsSummary;
+    getResultAndUsage() {
+        return { result: this.result, usage: this.usage };
     }
     getArtifactInputsFromServer() {
         return this.artifactsInputs;
@@ -37,9 +38,19 @@ class FileServiceServer {
         const app = (0, express_1.default)();
         app.use(express_1.default.json({ limit: "50mb" }));
         (0, ipc_1.humanLoopRoute)(app);
-        app.post("/test", async (req, res) => {
-            const { generatedCode, importPaths, actionsSummary } = req.body;
-            this.actionsSummary = actionsSummary;
+        app.post("/agent-results", async (req, res) => {
+            const { generatedCode, importPaths, result, usage } = req.body;
+            this.result = result;
+            this.usage = usage;
+            this.artifactsInputs = this.result
+                .filter((item) => item.type === "screenshot")
+                .map((item) => item.screenshot)
+                .map((artifact) => ({
+                ...artifact,
+                data: Buffer.isBuffer(artifact.data)
+                    ? artifact.data
+                    : Buffer.from(artifact.data),
+            }));
             if (!this.updateFile) {
                 // Not updating the file in this scenario
                 if (this.onComplete)
@@ -64,28 +75,6 @@ class FileServiceServer {
             }
             return res.send({ success: false });
         });
-        app.post("/artifact", async (req, res) => {
-            const artifactInputs = Array.isArray(req.body) ? req.body : [req.body];
-            for (const artifactInput of artifactInputs) {
-                const { name, contentType, data } = artifactInput;
-                if (!data) {
-                    console.error("No data in artifact", artifactInput);
-                    continue;
-                }
-                // Passing buffer data to the server
-                // Serialises the buffer data to an obbject
-                // Which needs to be converted back to a buffer
-                const bufferData = Buffer.isBuffer(data)
-                    ? data
-                    : Buffer.from(data.data);
-                this.artifactsInputs.push({
-                    name,
-                    contentType,
-                    data: bufferData,
-                });
-            }
-            return res.send({ success: true });
-        });
         return new Promise((resolve) => {
             this.server = app.listen(this.port, () => resolve(this.port));
         });

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,YAAY,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAQhD,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAqB7C,wBAAsB,UAAU,CAC9B,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,IAAI,GAAG,YAAY,EAC5B,KAAK,CAAC,EAAE,SAAS,iBA+ElB"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,YAAY,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAWhD,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAsB7C,wBAAsB,UAAU,CAC9B,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,IAAI,GAAG,YAAY,EAC5B,KAAK,CAAC,EAAE,SAAS,iBAwElB"}

package/dist/index.js

@@ -78,21 +78,14 @@ async function createTest(task, pageRef, scope) {
                 scopeVars: scope,
             });
         }
-        const { code, importPaths, actionsSummary, artifacts } = agentResult;
-        if (artifacts && artifacts.length > 0) {
-            await fileServiceClient.sendArtifactInputsToServer(artifacts);
-        }
-        await fileServiceClient.updateTest({
+        await fileServiceClient.sendAgentResult({
             task,
-            generatedCode: code,
-            importPaths,
-            actionsSummary,
-        });
-        span?.end({
-            output: {
-                actionsSummary,
-            },
+            result: agentResult.result,
+            usage: agentResult.usage,
+            generatedCode: agentResult.code,
+            importPaths: agentResult.importPaths,
         });
+        span?.end({ output: { agentResult } });
     }
     finally {
         // Ensure listeners are removed even if an error occurs

package/dist/tool-call-service/index.d.ts

@@ -1,7 +1,6 @@
 import { TraceClient } from "@empiricalrun/llm";
-import { PendingToolCall, SupportedChatModels, ToolResult } from "@empiricalrun/llm/chat";
-import { Tool, ToolExecute } from "@empiricalrun/shared-types";
-export type { SupportedChatModels };
+import { PendingToolCall, ToolResult } from "@empiricalrun/llm/chat";
+import { SupportedChatModels, Tool, ToolExecute } from "@empiricalrun/shared-types";
 type ToolExecutors = {
     [key: string]: ToolExecute;
 };
@@ -29,4 +28,5 @@ export declare class ToolCallService {
     sendToQueue(toolCalls: PendingToolCall[]): Promise<void>;
     execute(toolCalls: PendingToolCall[]): Promise<ToolResult[]>;
 }
+export {};
 //# sourceMappingURL=index.d.ts.map

package/dist/tool-call-service/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EAEL,eAAe,EACf,mBAAmB,EACnB,UAAU,EACX,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAiB,IAAI,EAAE,WAAW,EAAE,MAAM,4BAA4B,CAAC;AAiB9E,YAAY,EAAE,mBAAmB,EAAE,CAAC;AAEpC,KAAK,aAAa,GAAG;IACnB,CAAC,GAAG,EAAE,MAAM,GAAG,WAAW,CAAC;CAC5B,CAAC;AAEF,qBAAa,eAAe;IAC1B,KAAK,EAAE,IAAI,EAAE,CAAM;IACnB,aAAa,EAAE,aAAa,CAAM;IAClC,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,aAAa,EAAE,mBAAmB,CAAC;IACnC,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,oBAAoB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;gBACjC,EACV,aAAa,EACb,aAAa,EACb,UAAU,EACV,QAAQ,EACR,MAAM,EACN,KAAK,EACL,YAAY,EACZ,oBAAyB,GAC1B,EAAE;QACD,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;QAC7B,aAAa,EAAE,mBAAmB,CAAC;QACnC,UAAU,EAAE,MAAM,CAAC;QACnB,QAAQ,EAAE,MAAM,CAAC;QACjB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;QAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;QACpB,YAAY,EAAE,MAAM,EAAE,CAAC;QACvB,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;KAC/C;IAmCK,WAAW,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAkBxD,OAAO,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,UAAU,EAAE,CAAC;CAuEnE"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EAEL,eAAe,EACf,UAAU,EACX,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAEL,mBAAmB,EACnB,IAAI,EACJ,WAAW,EACZ,MAAM,4BAA4B,CAAC;AAmBpC,KAAK,aAAa,GAAG;IACnB,CAAC,GAAG,EAAE,MAAM,GAAG,WAAW,CAAC;CAC5B,CAAC;AAEF,qBAAa,eAAe;IAC1B,KAAK,EAAE,IAAI,EAAE,CAAM;IACnB,aAAa,EAAE,aAAa,CAAM;IAClC,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,aAAa,EAAE,mBAAmB,CAAC;IACnC,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,oBAAoB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;gBAEjC,EACV,aAAa,EACb,aAAa,EACb,UAAU,EACV,QAAQ,EACR,MAAM,EACN,KAAK,EACL,YAAY,EACZ,oBAAyB,GAC1B,EAAE;QACD,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;QAC7B,aAAa,EAAE,mBAAmB,CAAC;QACnC,UAAU,EAAE,MAAM,CAAC;QACnB,QAAQ,EAAE,MAAM,CAAC;QACjB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;QAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;QACpB,YAAY,EAAE,MAAM,EAAE,CAAC;QACvB,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;KAC/C;IAoCK,WAAW,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAkBxD,OAAO,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,UAAU,EAAE,CAAC;CAwEnE"}

package/dist/tool-call-service/index.js

@@ -6,12 +6,13 @@ const artifacts_1 = require("../artifacts");
 const commit_and_create_pr_1 = require("../tools/commit-and-create-pr");
 const diagnosis_fetcher_1 = require("../tools/diagnosis-fetcher");
 const download_build_1 = require("../tools/download-build");
-const environment_crud_1 = require("../tools/environment-crud");
 const grep_1 = require("../tools/grep");
+const list_environments_1 = require("../tools/list-environments");
 const str_replace_editor_1 = require("../tools/str_replace_editor");
 const test_gen_browser_1 = require("../tools/test-gen-browser");
 const test_run_1 = require("../tools/test-run");
 const test_run_fetcher_1 = require("../tools/test-run-fetcher");
+const upgrade_packages_1 = require("../tools/upgrade-packages");
 const checkpoint_1 = require("../utils/checkpoint");
 const utils_1 = require("./utils");
 class ToolCallService {
@@ -41,8 +42,9 @@ class ToolCallService {
             diagnosis_fetcher_1.fetchDiagnosisReportTool,
             test_gen_browser_1.generateTestWithBrowserAgent,
             commit_and_create_pr_1.createPullRequestTool,
-            environment_crud_1.getEnvironmentTool,
+            list_environments_1.listEnvironmentsTool,
             download_build_1.downloadBuildTool,
+            upgrade_packages_1.upgradePackagesTool,
         ];
         if ((0, chat_1.getProviderForModel)(this.selectedModel) !== "claude") {
             this.tools.push(...str_replace_editor_1.textEditorTools);
@@ -64,7 +66,7 @@ class ToolCallService {
         if (!requestId) {
             throw new Error("Could not find an id for the tool call.");
         }
-        const queueUrl = (0, utils_1.getQueueUrl)(this.featureFlags);
+        const queueUrl = (0, utils_1.getQueueUrl)(toolCalls, this.tools);
         if (!queueUrl) {
             throw new Error(`queueUrl is required for remote execution.`);
         }
@@ -116,6 +118,7 @@ class ToolCallService {
                     trace: this.trace,
                     collectArtifacts: collectArtifactsFn,
                     environmentOverrides: this.environmentOverrides,
+                    featureFlags: this.featureFlags,
                 });
                 const artifacts = await uploadArtifactsQueue.waitForCompletion();
                 toolResults.push({ ...result, artifacts });

package/dist/tool-call-service/utils.d.ts

@@ -1,11 +1,5 @@
-import { SupportedChatModels } from "@empiricalrun/llm/chat";
-import { PendingToolCall } from "@empiricalrun/shared-types";
-export declare function getQueueUrl(featureFlags: string[]): string | undefined;
-export declare function sendToolRequestToRemoteQueue(queueUrl: string, payload: {
-    toolCalls: PendingToolCall[];
-    requestId: string;
-    chatSessionId: number;
-    selectedModel: SupportedChatModels;
-    branchName: string;
-}): Promise<void>;
+import { PendingToolCall, ServicePayload, Tool } from "@empiricalrun/shared-types";
+export declare function appendBranchNameToQueueUrl(queueUrl: string, branchName: string): string;
+export declare function getQueueUrl(toolCalls: PendingToolCall[], tools: Tool[]): string;
+export declare function sendToolRequestToRemoteQueue(queueUrl: string, payload: ServicePayload): Promise<void>;
 //# sourceMappingURL=utils.d.ts.map