@empiricalrun/test-gen 0.58.0 → 0.60.0

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, }) {
+async function runChatAgent({ modelInput, chatSessionId, useDiskForChatState, initialPromptPath, withRetry, }) {
     const MODEL_MAPPING = {
         "claude-3-7": "claude-3-7-sonnet-20250219",
         "claude-3-5": "claude-3-5-sonnet-20241022",
@@ -71,6 +71,7 @@ async function runChatAgent({ modelInput, chatSessionId, useDiskForChatState, in
         selectedModel: specifiedModel || defaultModel,
         useDiskForChatState: useDiskForChatState || false,
         initialPromptContent,
+        withRetry,
     });
 }
 async function runAgentsWorkflow(testGenConfig, testGenToken) {
@@ -184,10 +185,17 @@ async function runAgentsWorkflow(testGenConfig, testGenToken) {
         void (0, session_1.updateSessionStatus)(testGenConfig.options?.metadata.testSessionId, {
             status: "agent_live_session_started",
         });
+        const projectName = await (0, run_1.convertProjectsFilterToProject)({
+            pwProjectsFilter: testGenConfig.environment?.playwrightProjects,
+            repoDir: process.cwd(),
+            testFilePath: specPath,
+        });
         const { isError, error } = await (0, run_1.generateTestsUsingMasterAgent)({
+            testCaseName: testCase.name,
+            testCaseSuites: testCase.suites,
             testFilePath: specPath,
             filePathToUpdate,
-            pwProjectsFilter: testGenConfig.environment?.playwrightProjects,
+            projectName,
             testGenToken,
             repoDir: process.cwd(),
             editFileWithGeneratedCode: true,
@@ -213,6 +221,7 @@ async function main() {
         .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("--initial-prompt <path>", "Path to an initial prompt file (e.g. prompt.md)")
+        .option("--with-retry", "Use the retry strategy")
         .parse(process.argv);
     const options = program.opts();
     const completedOptions = await (0, utils_2.validateAndCompleteCliOptions)(options);
@@ -243,6 +252,7 @@ async function main() {
             modelInput: completedOptions.chatModel,
             useDiskForChatState: completedOptions.useDiskForChatState,
             initialPromptPath: completedOptions.initialPrompt,
+            withRetry: completedOptions.withRetry,
         });
         return;
     }

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

@@ -9,6 +9,7 @@ export interface CliOptions {
     initialPrompt?: string;
     chatSessionId?: string;
     chatModel?: "claude-3-7" | "claude-3-5" | "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro" | "gemini-2.5-pro-preview-03-25" | "o4-mini" | "o4-mini-2025-04-16";
+    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":"AAEA,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,EACN,YAAY,GACZ,YAAY,GACZ,4BAA4B,GAC5B,4BAA4B,GAC5B,gBAAgB,GAChB,8BAA8B,GAC9B,SAAS,GACT,oBAAoB,CAAC;CAC1B;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":"AAEA,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,EACN,YAAY,GACZ,YAAY,GACZ,4BAA4B,GAC5B,4BAA4B,GAC5B,gBAAgB,GAChB,8BAA8B,GAC9B,SAAS,GACT,oBAAoB,CAAC;IACzB,SAAS,CAAC,EAAE,OAAO,CAAC;CACrB;AAQD,wBAAsB,6BAA6B,CACjD,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC,UAAU,CAAC,CAyDrB;AAED,wBAAgB,WAAW,SAgC1B"}

package/dist/index.d.ts

@@ -1,4 +1,5 @@
 import { FrameLocator, Page } from "playwright";
 import { ScopeVars } from "./types";
+export { downloadBuild } from "./test-build";
 export declare function createTest(task: string, pageRef: Page | FrameLocator, scope?: ScopeVars): Promise<void>;
 //# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,YAAY,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAQhD,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAoBpC,wBAAsB,UAAU,CAC9B,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,IAAI,GAAG,YAAY,EAC5B,KAAK,CAAC,EAAE,SAAS,iBA0ElB"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AACA,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;AAoB7C,wBAAsB,UAAU,CAC9B,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,IAAI,GAAG,YAAY,EAC5B,KAAK,CAAC,EAAE,SAAS,iBA0ElB"}

package/dist/index.js

@@ -3,6 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.downloadBuild = void 0;
 exports.createTest = createTest;
 const llm_1 = require("@empiricalrun/llm");
 const cua_1 = require("./agent/cua");
@@ -11,6 +12,8 @@ const scenarios_1 = require("./bin/utils/scenarios");
 const client_1 = __importDefault(require("./file/client"));
 const reporter_1 = require("./reporter");
 const session_1 = require("./session");
+var test_build_1 = require("./test-build");
+Object.defineProperty(exports, "downloadBuild", { enumerable: true, get: function () { return test_build_1.downloadBuild; } });
 const flushEvents = async () => {
     await (0, llm_1.flushAllTraces)();
 };

package/dist/test-build/index.js

@@ -23,7 +23,7 @@ async function downloadBuild(buildUrl) {
     const buildDownloadScript = packageJSON.scripts["download"];
     if (buildDownloadScript && buildUrl) {
         logger.log(`Downloading build from ${buildUrl}`);
-        await (0, exec_1.cmd)(`npm run download ${buildUrl}`.split(" "), {
+        await (0, exec_1.cmd)(`npm`, ["run", "download", buildUrl], {
             env: { ...Object(process.env) },
         });
     }

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

@@ -14,6 +14,7 @@ export declare class ToolCallService {
     getTools(): Promise<{
         tools: Tool[];
     }>;
-    execute(toolCalls: PendingToolCall[], isRemote: boolean, trace?: TraceClient): Promise<ToolResult[]>;
+    sendToQueue(toolCalls: PendingToolCall[]): Promise<void>;
+    execute(toolCalls: PendingToolCall[], trace?: TraceClient): Promise<ToolResult[]>;
 }
 //# 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":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EACL,eAAe,EACf,mBAAmB,EACnB,IAAI,EACJ,UAAU,EACX,MAAM,wBAAwB,CAAC;AAgBhC,YAAY,EAAE,mBAAmB,EAAE,CAAC;AAEpC,KAAK,aAAa,GAAG;IACnB,CAAC,GAAG,EAAE,MAAM,GAAG,CAAC,KAAK,EAAE,GAAG,EAAE,KAAK,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,UAAU,CAAC,CAAC;CACzE,CAAC;AA6BF,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;gBAEjB,aAAa,EAAE,MAAM,GAAG,IAAI,EAC5B,aAAa,EAAE,mBAAmB,EAClC,UAAU,EAAE,MAAM;IAiBd,QAAQ;;;IAaR,OAAO,CACX,SAAS,EAAE,eAAe,EAAE,EAC5B,QAAQ,EAAE,OAAO,EACjB,KAAK,CAAC,EAAE,WAAW,GAClB,OAAO,CAAC,UAAU,EAAE,CAAC;CAyDzB"}
+{"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,IAAI,EACJ,UAAU,EACX,MAAM,wBAAwB,CAAC;AAiBhC,YAAY,EAAE,mBAAmB,EAAE,CAAC;AAEpC,KAAK,aAAa,GAAG;IACnB,CAAC,GAAG,EAAE,MAAM,GAAG,CAAC,KAAK,EAAE,GAAG,EAAE,KAAK,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,UAAU,CAAC,CAAC;CACzE,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;gBAEjB,aAAa,EAAE,MAAM,GAAG,IAAI,EAC5B,aAAa,EAAE,mBAAmB,EAClC,UAAU,EAAE,MAAM;IAiBd,QAAQ;;;IAaR,WAAW,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAcxD,OAAO,CACX,SAAS,EAAE,eAAe,EAAE,EAC5B,KAAK,CAAC,EAAE,WAAW,GAClB,OAAO,CAAC,UAAU,EAAE,CAAC;CA2CzB"}

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

@@ -1,7 +1,7 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.ToolCallService = void 0;
-const client_sqs_1 = require("@aws-sdk/client-sqs");
+const chat_1 = require("@empiricalrun/llm/chat");
 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");
@@ -12,25 +12,7 @@ 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 checkpoint_1 = require("../utils/checkpoint");
-async function sendToolRequestToRemoteQueue(payload) {
-    const sqs = new client_sqs_1.SQSClient({
-        region: process.env.AWS_REGION,
-        credentials: {
-            accessKeyId: process.env.AWS_ACCESS_KEY_ID,
-            secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
-        },
-    });
-    const queueUrl = process.env.TOOL_EXECUTION_SQS_URL;
-    if (!queueUrl) {
-        throw new Error("TOOL_EXECUTION_SQS_URL is required for remote execution.");
-    }
-    await sqs.send(new client_sqs_1.SendMessageCommand({
-        QueueUrl: queueUrl,
-        MessageBody: JSON.stringify(payload),
-        MessageGroupId: payload.requestId,
-        MessageDeduplicationId: payload.requestId, // unique id for the tool request
-    }));
-}
+const utils_1 = require("./utils");
 class ToolCallService {
     tools = [];
     toolExecutors = {};
@@ -53,70 +35,68 @@ class ToolCallService {
         ];
     }
     async getTools() {
-        if (!this.selectedModel.startsWith("claude")) {
+        if ((0, chat_1.getProviderForModel)(this.selectedModel) !== "claude") {
             this.tools.push(...str_replace_editor_1.textEditorTools);
         }
         this.tools.forEach((tool) => {
             this.toolExecutors[tool.schema.name] = tool.execute;
         });
-        if (this.selectedModel.startsWith("claude")) {
+        if ((0, chat_1.getProviderForModel)(this.selectedModel) === "claude") {
             this.toolExecutors["str_replace_editor"] = str_replace_editor_1.strReplaceEditorExecutor;
         }
         return { tools: this.tools };
     }
-    async execute(toolCalls, isRemote, trace) {
-        if (isRemote && this.chatSessionId) {
-            await sendToolRequestToRemoteQueue({
-                toolCalls,
-                requestId: crypto.randomUUID(),
-                chatSessionId: this.chatSessionId,
-                selectedModel: this.selectedModel,
-                branchName: this.branchName,
-            });
-            return toolCalls.map(() => ({
-                isError: false,
-                result: `Tool request sent to remote queue to execute.`,
-            }));
+    async sendToQueue(toolCalls) {
+        const requestId = toolCalls[0]?.id;
+        if (!requestId) {
+            throw new Error("Could not find an id for the tool call.");
         }
-        else {
-            const executeSpan = trace?.span({
-                name: "execute_tools",
-                input: { toolCalls: toolCalls.map((tc) => ({ name: tc.name })) },
+        await (0, utils_1.sendToolRequestToRemoteQueue)({
+            toolCalls,
+            requestId,
+            chatSessionId: this.chatSessionId,
+            selectedModel: this.selectedModel,
+            branchName: this.branchName,
+        });
+    }
+    async execute(toolCalls, trace) {
+        const executeSpan = trace?.span({
+            name: "execute_tools",
+            input: { toolCalls: toolCalls.map((tc) => ({ name: tc.name })) },
+        });
+        const toolResults = [];
+        for (const toolCall of toolCalls) {
+            const span = executeSpan?.span({
+                name: `tool: ${toolCall.name}`,
+                input: toolCall.input,
             });
-            const toolResults = [];
-            for (const toolCall of toolCalls) {
-                const span = executeSpan?.span({
-                    name: `tool: ${toolCall.name}`,
-                    input: toolCall.input,
-                });
-                const toolExecutor = this.toolExecutors[toolCall.name];
-                if (!toolExecutor) {
-                    const errorResult = {
-                        isError: true,
-                        result: `Invalid function/tool call: invalid_tool_call not found`,
-                    };
-                    toolResults.push(errorResult);
-                    span?.end({ output: errorResult });
-                    continue;
-                }
-                try {
-                    const result = await toolExecutor(toolCall.input, trace);
-                    toolResults.push(result);
-                    span?.end({ output: result });
-                }
-                catch (error) {
-                    const errorResult = {
-                        isError: true,
-                        result: error instanceof Error ? error.message : String(error),
-                    };
-                    toolResults.push(errorResult);
-                    span?.end({ output: errorResult });
-                }
+            const toolExecutor = this.toolExecutors[toolCall.name];
+            if (!toolExecutor) {
+                const errorResult = {
+                    isError: true,
+                    result: `Invalid function/tool call: ${toolCall.name} not found`,
+                };
+                toolResults.push(errorResult);
+                span?.end({ output: errorResult });
+                continue;
+            }
+            try {
+                const result = await toolExecutor(toolCall.input, trace);
+                toolResults.push(result);
+                span?.end({ output: result });
+            }
+            catch (error) {
+                const errorResult = {
+                    isError: true,
+                    result: error instanceof Error ? error.message : String(error),
+                };
+                toolResults.push(errorResult);
+                span?.end({ output: errorResult });
             }
-            await (0, checkpoint_1.createCheckpoint)(toolCalls, this.branchName);
-            executeSpan?.end({ output: { toolResults } });
-            return toolResults;
         }
+        await (0, checkpoint_1.createCheckpoint)(toolCalls, this.branchName);
+        executeSpan?.end({ output: { toolResults } });
+        return toolResults;
     }
 }
 exports.ToolCallService = ToolCallService;

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

@@ -0,0 +1,10 @@
+import { SupportedChatModels } from "@empiricalrun/llm/chat";
+import { PendingToolCall } from "@empiricalrun/shared-types";
+export declare function sendToolRequestToRemoteQueue(payload: {
+    toolCalls: PendingToolCall[];
+    requestId: string;
+    chatSessionId: number;
+    selectedModel: SupportedChatModels;
+    branchName: string;
+}): Promise<void>;
+//# sourceMappingURL=utils.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/utils.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AAC7D,OAAO,EAAE,eAAe,EAAE,MAAM,4BAA4B,CAAC;AAE7D,wBAAsB,4BAA4B,CAAC,OAAO,EAAE;IAC1D,SAAS,EAAE,eAAe,EAAE,CAAC;IAC7B,SAAS,EAAE,MAAM,CAAC;IAClB,aAAa,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,mBAAmB,CAAC;IACnC,UAAU,EAAE,MAAM,CAAC;CACpB,iBAoBA"}

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

@@ -0,0 +1,23 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.sendToolRequestToRemoteQueue = sendToolRequestToRemoteQueue;
+const client_sqs_1 = require("@aws-sdk/client-sqs");
+async function sendToolRequestToRemoteQueue(payload) {
+    const sqs = new client_sqs_1.SQSClient({
+        region: process.env.AWS_REGION,
+        credentials: {
+            accessKeyId: process.env.AWS_ACCESS_KEY_ID,
+            secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
+        },
+    });
+    const queueUrl = process.env.TOOL_EXECUTION_SQS_URL;
+    if (!queueUrl) {
+        throw new Error("TOOL_EXECUTION_SQS_URL is required for remote execution.");
+    }
+    await sqs.send(new client_sqs_1.SendMessageCommand({
+        QueueUrl: queueUrl,
+        MessageBody: JSON.stringify(payload),
+        MessageGroupId: payload.requestId,
+        MessageDeduplicationId: payload.requestId, // unique id for the tool request
+    }));
+}

package/dist/tools/download-build.d.ts

@@ -1,3 +1,12 @@
 import type { Tool } from "@empiricalrun/llm/chat";
+import { z } from "zod";
+export declare const downloadBuildToolSchema: z.ZodObject<{
+    buildUrl: z.ZodString;
+}, "strip", z.ZodTypeAny, {
+    buildUrl: string;
+}, {
+    buildUrl: string;
+}>;
+export type DownloadBuildToolInput = z.infer<typeof downloadBuildToolSchema>;
 export declare const downloadBuildTool: Tool;
 //# sourceMappingURL=download-build.d.ts.map

package/dist/tools/download-build.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"download-build.d.ts","sourceRoot":"","sources":["../../src/tools/download-build.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAKnD,eAAO,MAAM,iBAAiB,EAAE,IAkC/B,CAAC"}
+{"version":3,"file":"download-build.d.ts","sourceRoot":"","sources":["../../src/tools/download-build.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AACnD,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,uBAAuB;;;;;;EAElC,CAAC;AAEH,MAAM,MAAM,sBAAsB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AAE7E,eAAO,MAAM,iBAAiB,EAAE,IA8B/B,CAAC"}

package/dist/tools/download-build.js

@@ -1,17 +1,18 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.downloadBuildTool = void 0;
+exports.downloadBuildTool = exports.downloadBuildToolSchema = void 0;
 const zod_1 = require("zod");
 const test_build_1 = require("../test-build");
+exports.downloadBuildToolSchema = zod_1.z.object({
+    buildUrl: zod_1.z.string().describe("The URL of the build to download"),
+});
 exports.downloadBuildTool = {
     schema: {
         name: "downloadBuild",
         description: `Download a build from a build URL. If you do not have
 have a build URL, you can try getting the environment details with the getEnvironment tool.
 Environment details will include the build URL.`,
-        parameters: zod_1.z.object({
-            buildUrl: zod_1.z.string().describe("The URL of the build to download"),
-        }),
+        parameters: exports.downloadBuildToolSchema,
     },
     execute: async (input) => {
         if (!(await (0, test_build_1.hasDownloadScript)())) {

package/dist/tools/str_replace_editor.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"str_replace_editor.d.ts","sourceRoot":"","sources":["../../src/tools/str_replace_editor.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AA2B1D,UAAU,eAAe;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAqED;;;GAGG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAwC1D;AAMD;;;GAGG;AACH,wBAAsB,wBAAwB,CAC5C,KAAK,EAAE,eAAe,GACrB,OAAO,CAAC,UAAU,CAAC,CA0KrB;AA+FD,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}
+{"version":3,"file":"str_replace_editor.d.ts","sourceRoot":"","sources":["../../src/tools/str_replace_editor.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AA2B1D,UAAU,eAAe;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAqED;;;GAGG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAwC1D;AAMD;;;GAGG;AACH,wBAAsB,wBAAwB,CAC5C,KAAK,EAAE,eAAe,GACrB,OAAO,CAAC,UAAU,CAAC,CA6LrB;AAiGD,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}

package/dist/tools/str_replace_editor.js

@@ -148,7 +148,6 @@ async function strReplaceEditorExecutor(input) {
         let typeCheckErrors;
         switch (input.command) {
             case "view":
-                // TODO: This assumes repoDir is process.cwd()
                 if (!fs_1.default.existsSync(filePath)) {
                     return {
                         result: "Error: File not found",
@@ -163,6 +162,17 @@ async function strReplaceEditorExecutor(input) {
                         isError: false,
                     };
                 }
+                else {
+                    // Check if file is binary, which is not supported
+                    const { isBinary } = await import("istextorbinary");
+                    const binary = isBinary(filePath);
+                    if (binary) {
+                        return {
+                            result: "Error: File is binary, which is not supported",
+                            isError: true,
+                        };
+                    }
+                }
                 // Handle file view
                 content = fs_1.default.readFileSync(filePath, "utf8");
                 lines = content.split("\n");
@@ -254,7 +264,13 @@ async function strReplaceEditorExecutor(input) {
                 createBackup(filePath);
                 content = fs_1.default.readFileSync(filePath, "utf8");
                 lines = content.split("\n");
-                lines.splice(input.insert_line, 0, input.new_str);
+                if (input.insert_line < 1) {
+                    throw new Error("insert_line must be greater than or equal to 1 (line numbers are 1-indexed).");
+                }
+                if (input.insert_line > lines.length + 1) {
+                    throw new Error(`The file at ${filePath} has only ${lines.length} lines, so insert_line must be less than or equal to ${lines.length + 1}. At the maximum value of ${lines.length + 1}, you can insert at the end of the file.`);
+                }
+                lines.splice(input.insert_line - 1, 0, input.new_str);
                 fs_1.default.writeFileSync(filePath, lines.join("\n"));
                 typeCheckErrors = (0, web_1.validateTypescript)(filePath);
                 if (typeCheckErrors.length > 0) {
@@ -297,8 +313,9 @@ const fileViewTool = {
         name: "fileViewTool",
         description: `A tool to view the content of a file or directory. If the path points 
 to a directory, the tool will return a list of files in the directory, separated by line breaks.
-If the path points to a file, the tool will return the content of the file. File contents
-are returned with line numbers, starting from 1.
+If the path points to a file, the tool will return the content of the file.
+
+File contents are returned with line numbers, starting from 1.
 
 1: line 1
 2: line 2
@@ -355,14 +372,14 @@ in the file. If old_str is not unique, the tool will return an error.`,
 const stringInsertTool = {
     schema: {
         name: "stringInsertTool",
-        description: "A tool to insert a string at a specific line in a file.",
+        description: "A tool to insert a string at a specific line in a file. Line numbers are 1-indexed, just like the file view tool.",
         parameters: zod_1.z.object({
             path: zod_1.z.string().describe("The path to the file."),
             insert_line: zod_1.z
                 .number()
                 .int()
-                .min(0)
-                .describe("The line number after which to insert the text (0 for beginning of file)."),
+                .min(1)
+                .describe("The line number after which to insert the text (1 for beginning of file)."),
             new_str: zod_1.z.string().describe("The string to insert."),
         }),
     },

package/dist/tools/test-gen-browser.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"test-gen-browser.d.ts","sourceRoot":"","sources":["../../src/tools/test-gen-browser.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAyFnD,eAAO,MAAM,4BAA4B,EAAE,IA8E1C,CAAC"}
+{"version":3,"file":"test-gen-browser.d.ts","sourceRoot":"","sources":["../../src/tools/test-gen-browser.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAsFnD,eAAO,MAAM,4BAA4B,EAAE,IAqF1C,CAAC"}

package/dist/tools/test-gen-browser.js

@@ -4,6 +4,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
 };
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.generateTestWithBrowserAgent = void 0;
+const test_run_1 = require("@empiricalrun/test-run");
 const promises_1 = __importDefault(require("fs/promises"));
 const zod_1 = require("zod");
 const run_1 = require("../agent/browsing/run");
@@ -87,13 +88,23 @@ exports.generateTestWithBrowserAgent = {
         parameters: BrowserAgentSchema,
     },
     execute: async (input, trace) => {
+        const repoDir = process.cwd();
         const { testName, testSuites, fileName, changeToMake, project } = input;
-        const playwrightConfig = await (0, utils_1.readPlaywrightConfig)(process.cwd());
-        const validProjectNames = await (0, utils_1.getValidProjectNames)(playwrightConfig);
-        if (!validProjectNames.includes(project)) {
+        try {
+            const { projects } = await (0, test_run_1.getAllPlaywrightProjects)(repoDir);
+            // TODO: Check that file path is valid for this project
+            if (!projects.includes(project)) {
+                return {
+                    isError: true,
+                    result: `Invalid project name: ${project}. Valid project names are: ${projects.join(", ")}`,
+                };
+            }
+        }
+        catch (error) {
+            console.error("Error reading playwright config:", error);
             return {
                 isError: true,
-                result: `Invalid project name: ${project}. Valid project names are: ${validProjectNames.join(", ")}`,
+                result: `Error reading playwright config: ${error}`,
             };
         }
         if (!(0, web_1.hasTestBlock)({ testName, testSuites, filePath: fileName })) {
@@ -104,34 +115,30 @@ exports.generateTestWithBrowserAgent = {
         }
         const fileBackup = await promises_1.default.readFile(fileName, "utf-8");
         try {
-            await (0, utils_1.replaceTodoWithCreateTest)({
-                testCaseName: testName,
-                testCaseSuites: testSuites,
-                testFilePath: fileName,
-            });
+            await (0, utils_1.replaceTodoWithCreateTest)(fileName);
         }
         catch (error) {
+            // Undo the TODO -> createTest and test.only changes
+            await promises_1.default.writeFile(fileName, fileBackup, "utf-8");
             return {
                 isError: true,
                 result: `Error running tool: ${error}`,
             };
         }
+        const testGenToken = (0, scenarios_1.buildTokenFromOptions)({ name: testName, file: fileName, prompt: changeToMake }, { useComputerUseAgent: true });
+        console.log("[generateTestWithBrowserAgent] Validations passed, starting agent");
         const toolResult = await (0, run_1.generateTestsUsingMasterAgent)({
+            testCaseName: testName,
+            testCaseSuites: testSuites,
             testFilePath: fileName,
             filePathToUpdate: fileName,
-            pwProjectsFilter: [project],
+            projectName: project,
             traceId: trace?.id,
-            testGenToken: (0, scenarios_1.buildTokenFromOptions)({
-                name: testName,
-                file: fileName,
-                prompt: changeToMake,
-            }, {
-                useComputerUseAgent: true,
-            }),
-            repoDir: process.cwd(),
+            testGenToken,
+            repoDir,
             editFileWithGeneratedCode: false,
         });
-        // Undo the TODO -> createTest change
+        // Undo the TODO -> createTest changes
         await promises_1.default.writeFile(fileName, fileBackup, "utf-8");
         const { isError, error, actionsSummary } = toolResult;
         if (!isError) {

package/dist/tools/test-run.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"test-run.d.ts","sourceRoot":"","sources":["../../src/tools/test-run.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAgDnD,eAAO,MAAM,WAAW,EAAE,IAuDzB,CAAC"}
+{"version":3,"file":"test-run.d.ts","sourceRoot":"","sources":["../../src/tools/test-run.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAgDnD,eAAO,MAAM,WAAW,EAAE,IAiDzB,CAAC"}

package/dist/tools/test-run.js

@@ -39,32 +39,27 @@ exports.runTestTool = {
     },
     execute: async (input) => {
         let reportUrl = undefined;
-        let projectName = undefined;
-        let testRunId = undefined;
+        let envOverrides = undefined;
         if (hasCloudflareCredentials()) {
-            projectName = "test-gen-chat-agent";
-            testRunId = Date.now().toString();
+            const projectName = "test-gen-chat-agent";
+            const testRunId = Date.now().toString();
             reportUrl = buildReportUrl(projectName, testRunId);
+            envOverrides = {
+                PROJECT_NAME: projectName,
+                TEST_RUN_GITHUB_ACTION_ID: testRunId,
+            };
         }
         else {
             console.warn("R2 credentials not found: report artifacts will not be uploaded");
         }
         const { testName, suites, fileName, project } = input;
         try {
-            // {"project":"chromium","suites":[],"fileName":"tests/quizizz-for-work/group.spec.ts","testName":"Create a group"}
-            // This runs all tests - TODO: Debug this, should only run the testName
             const result = await (0, test_run_1.runSingleTest)({
                 testName,
                 suites,
                 fileName,
                 projects: [project],
-                // Adding these to enforce report artifacts are uploaded
-                envOverrides: projectName && testRunId
-                    ? {
-                        PROJECT_NAME: projectName,
-                        TEST_RUN_GITHUB_ACTION_ID: testRunId,
-                    }
-                    : undefined,
+                envOverrides,
             });
             return {
                 result: buildResult({

package/dist/utils/checkpoint.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"checkpoint.d.ts","sourceRoot":"","sources":["../../src/utils/checkpoint.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AAIzD,wBAAsB,gBAAgB,CACpC,SAAS,EAAE,eAAe,EAAE,EAC5B,UAAU,EAAE,MAAM,iBAcnB"}
+{"version":3,"file":"checkpoint.d.ts","sourceRoot":"","sources":["../../src/utils/checkpoint.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AAIzD,wBAAsB,gBAAgB,CACpC,SAAS,EAAE,eAAe,EAAE,EAC5B,UAAU,EAAE,MAAM,iBAqBnB"}

package/dist/utils/checkpoint.js

@@ -5,7 +5,9 @@ const git_1 = require("./git");
 async function createCheckpoint(toolCalls, branchName) {
     const filesChanged = await (0, git_1.getFilesChanged)();
     const toolsWithUpdatedFiles = toolCalls
-        .filter((tc) => tc.input.path && filesChanged.includes(tc.input.path))
+        .filter((tc) => "path" in tc.input &&
+        tc.input.path &&
+        filesChanged.includes(tc.input.path))
         .map((toolCall) => ({
         name: toolCall.name,
         path: toolCall.input.path,

package/dist/utils/exec.d.ts

@@ -1,12 +1,12 @@
 export declare class ProcessManager {
     private childProcess;
-    execute(command: string[], options: {
+    execute(command: string, args: string[], options: {
         env?: Record<string, string>;
     }): Promise<number>;
     terminate(): void;
     isRunning(): boolean;
 }
-export declare function cmd(command: string[], options: {
+export declare function cmd(command: string, args: string[], options: {
     env?: Record<string, string>;
 }): Promise<number>;
 //# sourceMappingURL=exec.d.ts.map

package/dist/utils/exec.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"exec.d.ts","sourceRoot":"","sources":["../../src/utils/exec.ts"],"names":[],"mappings":"AAmBA,qBAAa,cAAc;IACzB,OAAO,CAAC,YAAY,CAA6B;IAE3C,OAAO,CACX,OAAO,EAAE,MAAM,EAAE,EACjB,OAAO,EAAE;QAAE,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,GACxC,OAAO,CAAC,MAAM,CAAC;IAkDlB,SAAS,IAAI,IAAI;IASjB,SAAS,IAAI,OAAO;CAGrB;AAED,wBAAsB,GAAG,CACvB,OAAO,EAAE,MAAM,EAAE,EACjB,OAAO,EAAE;IAAE,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CAAE,GACxC,OAAO,CAAC,MAAM,CAAC,CAGjB"}
+{"version":3,"file":"exec.d.ts","sourceRoot":"","sources":["../../src/utils/exec.ts"],"names":[],"mappings":"AAmBA,qBAAa,cAAc;IACzB,OAAO,CAAC,YAAY,CAA6B;IAE3C,OAAO,CACX,OAAO,EAAE,MAAM,EACf,IAAI,EAAE,MAAM,EAAE,EACd,OAAO,EAAE;QAAE,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,GACxC,OAAO,CAAC,MAAM,CAAC;IAmDlB,SAAS,IAAI,IAAI;IASjB,SAAS,IAAI,OAAO;CAGrB;AAED,wBAAsB,GAAG,CACvB,OAAO,EAAE,MAAM,EACf,IAAI,EAAE,MAAM,EAAE,EACd,OAAO,EAAE;IAAE,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CAAE,GACxC,OAAO,CAAC,MAAM,CAAC,CAGjB"}