@empiricalrun/test-gen 0.61.0 → 0.63.0

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

@@ -2,6 +2,7 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.ToolCallService = void 0;
 const chat_1 = require("@empiricalrun/llm/chat");
+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");
@@ -19,10 +20,16 @@ class ToolCallService {
     chatSessionId;
     selectedModel;
     branchName;
-    constructor(chatSessionId, selectedModel, branchName) {
+    repoPath;
+    apiKey;
+    trace;
+    constructor({ chatSessionId, selectedModel, branchName, repoPath, apiKey, trace, }) {
         this.chatSessionId = chatSessionId;
         this.selectedModel = selectedModel;
         this.branchName = branchName;
+        this.trace = trace;
+        this.repoPath = repoPath;
+        this.apiKey = apiKey;
         this.tools = [
             grep_1.grepTool,
             test_run_1.runTestTool,
@@ -33,8 +40,6 @@ class ToolCallService {
             environment_crud_1.getEnvironmentTool,
             download_build_1.downloadBuildTool,
         ];
-    }
-    async getTools() {
         if ((0, chat_1.getProviderForModel)(this.selectedModel) !== "claude") {
             this.tools.push(...str_replace_editor_1.textEditorTools);
         }
@@ -42,9 +47,13 @@ class ToolCallService {
             this.toolExecutors[tool.schema.name] = tool.execute;
         });
         if ((0, chat_1.getProviderForModel)(this.selectedModel) === "claude") {
-            this.toolExecutors["str_replace_editor"] = str_replace_editor_1.strReplaceEditorExecutor;
+            this.toolExecutors = {
+                ...this.toolExecutors,
+                // Support for Claude 3x and 4: They use different tool names
+                str_replace_editor: str_replace_editor_1.strReplaceEditorExecutor,
+                str_replace_based_edit_tool: str_replace_editor_1.strReplaceEditorExecutor,
+            };
         }
-        return { tools: this.tools };
     }
     async sendToQueue(toolCalls) {
         const requestId = toolCalls[0]?.id;
@@ -59,8 +68,8 @@ class ToolCallService {
             branchName: this.branchName,
         });
     }
-    async execute(toolCalls, trace) {
-        const executeSpan = trace?.span({
+    async execute(toolCalls) {
+        const executeSpan = this.trace?.span({
             name: "execute_tools",
             input: { toolCalls: toolCalls.map((tc) => ({ name: tc.name })) },
         });
@@ -70,31 +79,48 @@ class ToolCallService {
                 name: `tool: ${toolCall.name}`,
                 input: toolCall.input,
             });
+            const uploadArtifactsQueue = new artifacts_1.UploadArtifactsQueue(this.repoPath, toolCall.id);
+            const collectArtifactsFn = (artifactsInput) => {
+                uploadArtifactsQueue.addTask(artifactsInput).catch((error) => {
+                    console.error("Error collecting artifacts:", error);
+                });
+                return;
+            };
             const toolExecutor = this.toolExecutors[toolCall.name];
             if (!toolExecutor) {
                 const errorResult = {
                     isError: true,
                     result: `Invalid function/tool call: ${toolCall.name} not found`,
+                    artifacts: null,
                 };
                 toolResults.push(errorResult);
                 span?.end({ output: errorResult });
                 continue;
             }
+            if (!this.apiKey) {
+                throw new Error("API key is required for tool execution");
+            }
             try {
-                const result = await toolExecutor(toolCall.input, trace);
-                toolResults.push(result);
+                const result = await toolExecutor(toolCall.input, this.repoPath, this.apiKey, this.trace, collectArtifactsFn);
+                const artifacts = await uploadArtifactsQueue.waitForCompletion();
+                toolResults.push({ ...result, artifacts });
                 span?.end({ output: result });
             }
             catch (error) {
                 const errorResult = {
                     isError: true,
                     result: error instanceof Error ? error.message : String(error),
+                    artifacts: null,
                 };
                 toolResults.push(errorResult);
                 span?.end({ output: errorResult });
             }
         }
-        await (0, checkpoint_1.createCheckpoint)(toolCalls, this.branchName);
+        await (0, checkpoint_1.createCheckpoint)({
+            toolCalls,
+            branchName: this.branchName,
+            repoPath: this.repoPath,
+        });
         executeSpan?.end({ output: { toolResults } });
         return toolResults;
     }

package/dist/tools/commit-and-create-pr.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"commit-and-create-pr.d.ts","sourceRoot":"","sources":["../../src/tools/commit-and-create-pr.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAOnD,UAAU,cAAc;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;CACd;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,GAAG,cAAc,CAiB1D;AA2CD,eAAO,MAAM,qBAAqB,EAAE,IA2EnC,CAAC"}
+{"version":3,"file":"commit-and-create-pr.d.ts","sourceRoot":"","sources":["../../src/tools/commit-and-create-pr.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAOnD,UAAU,cAAc;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;CACd;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,GAAG,cAAc,CAiB1D;AA2CD,eAAO,MAAM,qBAAqB,EAAE,IAsFnC,CAAC"}

package/dist/tools/commit-and-create-pr.js

@@ -55,11 +55,13 @@ Don't ask the user for this information, just come up with it yourself.
 `,
         parameters: createPullRequestSchema,
     },
-    execute: async (input) => {
+    execute: async (input, repoPath, apiKey) => {
         try {
             const { pullRequestTitle, pullRequestDescription } = input;
-            const branchName = await (0, git_1.getCurrentBranchName)();
-            const repoUrl = (0, child_process_1.execSync)("git config --get remote.origin.url")
+            const branchName = await (0, git_1.getCurrentBranchName)(repoPath);
+            const repoUrl = (0, child_process_1.execSync)("git config --get remote.origin.url", {
+                cwd: repoPath,
+            })
                 .toString()
                 .trim();
             const { owner, repo } = parseGitHubUrl(repoUrl);
@@ -70,8 +72,11 @@ Don't ask the user for this information, just come up with it yourself.
                     head: `${owner}:${branchName}`,
                     state: "open",
                 },
+                apiKey,
             }));
-            (0, child_process_1.execSync)(`git push origin ${branchName} --set-upstream`);
+            (0, child_process_1.execSync)(`git push origin ${branchName} --set-upstream`, {
+                cwd: repoPath,
+            });
             const existingPR = existingPRs?.find((pr) => pr.head.ref === branchName);
             if (existingPR) {
                 // Append the new description to the existing PR description
@@ -82,6 +87,7 @@ Don't ask the user for this information, just come up with it yourself.
                     body: {
                         body: updatedDescription,
                     },
+                    apiKey,
                 });
                 return {
                     isError: false,
@@ -98,6 +104,7 @@ Don't ask the user for this information, just come up with it yourself.
                     base: "main",
                     body: initialDescription,
                 },
+                apiKey,
             }));
             return {
                 isError: false,

package/dist/tools/diagnosis-fetcher.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"diagnosis-fetcher.d.ts","sourceRoot":"","sources":["../../src/tools/diagnosis-fetcher.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAenD,eAAO,MAAM,wBAAwB,EAAE,IAgFtC,CAAC"}
+{"version":3,"file":"diagnosis-fetcher.d.ts","sourceRoot":"","sources":["../../src/tools/diagnosis-fetcher.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAenD,eAAO,MAAM,wBAAwB,EAAE,IAqFtC,CAAC"}

package/dist/tools/diagnosis-fetcher.js

@@ -19,7 +19,7 @@ exports.fetchDiagnosisReportTool = {
         description: "Fetch details about a test case diagnosis using its URL or slug",
         parameters: DiagnosisSchema,
     },
-    execute: async (input) => {
+    execute: async (input, repoPath, apiKey) => {
         const { diagnosisUrl } = input;
         // Extract the slug from the URL - it's the part after the last '--'
         const slug = diagnosisUrl.split("--").pop();
@@ -33,6 +33,7 @@ exports.fetchDiagnosisReportTool = {
         try {
             data = await (0, utils_1.makeDashboardRequest)({
                 path: `/api/diagnosis/${slug}/detailed`,
+                apiKey,
             });
         }
         catch (error) {
@@ -43,8 +44,8 @@ exports.fetchDiagnosisReportTool = {
         }
         const { test_case, diagnosis } = data.data;
         const project = diagnosis?.test_project || "unknown";
-        const sourceContext = await promises_1.default.readFile(path_1.default.join("tests", test_case.file_path), "utf-8");
-        const repoName = path_1.default.basename(process.cwd());
+        const sourceContext = await promises_1.default.readFile(path_1.default.join(repoPath, "tests", test_case.file_path), "utf-8");
+        const repoName = path_1.default.basename(repoPath);
         const cleanErrorStack = diagnosis?.failed_run_metadata?.stack?.replace(`"/runner/_work/${repoName}/${repoName}/source-repo/"`, "");
         // Format the response as markdown
         const markdownResponse = `

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;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"}
+{"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,IAkC/B,CAAC"}

package/dist/tools/download-build.js

@@ -14,8 +14,8 @@ have a build URL, you can try getting the environment details with the getEnviro
 Environment details will include the build URL.`,
         parameters: exports.downloadBuildToolSchema,
     },
-    execute: async (input) => {
-        if (!(await (0, test_build_1.hasDownloadScript)())) {
+    execute: async (input, repoPath, apiKey) => {
+        if (!(await (0, test_build_1.hasDownloadScript)(repoPath))) {
             return {
                 isError: true,
                 result: `This repo does not have a download script in package.json.
@@ -24,7 +24,7 @@ You probably don't need to worry about this, since it means this repo does not h
         }
         const { buildUrl } = input;
         try {
-            await (0, test_build_1.downloadBuild)(buildUrl);
+            await (0, test_build_1.downloadBuild)({ buildUrl, repoPath, apiKey });
             return {
                 isError: false,
                 result: "Build downloaded successfully",

package/dist/tools/environment-crud.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"environment-crud.d.ts","sourceRoot":"","sources":["../../src/tools/environment-crud.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAyDnD,eAAO,MAAM,kBAAkB,EAAE,IAwEhC,CAAC;AAGF,eAAO,MAAM,gBAAgB,EAAE,IAAI,EAAyB,CAAC"}
+{"version":3,"file":"environment-crud.d.ts","sourceRoot":"","sources":["../../src/tools/environment-crud.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AA6DnD,eAAO,MAAM,kBAAkB,EAAE,IA6EhC,CAAC;AAGF,eAAO,MAAM,gBAAgB,EAAE,IAAI,EAAyB,CAAC"}

package/dist/tools/environment-crud.js

@@ -5,10 +5,11 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.environmentTools = exports.getEnvironmentTool = void 0;
 const fs_1 = __importDefault(require("fs"));
+const path_1 = __importDefault(require("path"));
 const zod_1 = require("zod");
 const utils_1 = require("./utils");
-const getProjectRepoName = () => {
-    const packageJson = fs_1.default.readFileSync("package.json", "utf8");
+const getProjectRepoName = (repoPath) => {
+    const packageJson = fs_1.default.readFileSync(path_1.default.join(repoPath, "package.json"), "utf8");
     if (!packageJson) {
         throw new Error("Could not find or read package.json file");
     }
@@ -30,11 +31,11 @@ exports.getEnvironmentTool = {
         description: "Fetch details of an existing environment",
         parameters: GetEnvironmentSchema,
     },
-    execute: async (input) => {
+    execute: async (input, repoPath, apiKey) => {
         // Get project repo name
         let projectRepoName;
         try {
-            projectRepoName = getProjectRepoName();
+            projectRepoName = getProjectRepoName(repoPath);
         }
         catch (error) {
             return {
@@ -52,6 +53,7 @@ exports.getEnvironmentTool = {
             response = await (0, utils_1.makeDashboardRequest)({
                 path: `/api/environments?${queryParams.toString()}`,
                 method: "GET",
+                apiKey,
             });
         }
         catch (error) {

package/dist/tools/grep/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/grep/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAc,MAAM,wBAAwB,CAAC;AAyH/D,eAAO,MAAM,QAAQ,EAAE,IAmBtB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/grep/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAc,MAAM,wBAAwB,CAAC;AA6H/D,eAAO,MAAM,QAAQ,EAAE,IAoBtB,CAAC"}

package/dist/tools/grep/index.js

@@ -23,9 +23,9 @@ const GrepInputSchema = zod_1.z.object({
         .optional()
         .describe("File pattern to search in (e.g., '*.ts' for TypeScript files)"),
 });
-async function usingSystemGrep(input) {
+async function usingSystemGrep(input, repoPath) {
     try {
-        const dir = input.directory || process.cwd();
+        const dir = path_1.default.join(repoPath, input.directory || "");
         // Create exclude pattern for grep
         const excludePatterns = repo_tree_1.DEFAULT_EXCLUDE.map((pattern) => typeof pattern === "string" ? pattern : pattern.source)
             .map((pattern) => `--exclude-dir="${pattern}"`)
@@ -71,11 +71,13 @@ async function usingSystemGrep(input) {
         };
     }
 }
-async function usingRipgrep(input) {
+async function usingRipgrep(input, repoPath) {
     try {
-        const dir = path_1.default.join(process.cwd(), input.directory || "");
-        const escapedPattern = input.pattern.replace(/\s+/g, "\\ ");
-        const results = await (0, ripgrep_1.ripgrep)(dir, {
+        const dir = path_1.default.join(repoPath, input.directory || "");
+        const escapedPattern = input.pattern
+            .replace(/\s+/g, "\\ ")
+            .replace(/([[\]()])/g, "\\$1");
+        const results = (0, ripgrep_1.ripgrep)(dir, {
             string: escapedPattern,
             globs: input.filePattern ? [input.filePattern] : undefined,
         });
@@ -84,7 +86,7 @@ async function usingRipgrep(input) {
             // Can add submatches and offset info to the summary if needed
             return {
                 lines: result.lines.text,
-                path: path_1.default.relative(process.cwd(), result.path.text),
+                path: path_1.default.relative(repoPath, result.path.text),
                 // line number is 1-indexed
                 line_number: result.line_number,
             };
@@ -96,7 +98,7 @@ ${result.path}:${result.line_number}
 ${result.lines}\`\`\`
 `;
         });
-        const relDir = path_1.default.relative(process.cwd(), dir);
+        const relDir = path_1.default.relative(repoPath, dir);
         const header = `Found ${resultsSummary.length} results for "${input.pattern}" in "${relDir}".
 All paths are relative to the current working directory.`;
         return {
@@ -121,13 +123,13 @@ If ripgrep is not available, it will fall back to using system grep.
 Search is case insensitive and regex patterns are not supported.`,
         parameters: GrepInputSchema,
     },
-    execute: async (input) => {
+    execute: async (input, repoPath) => {
         if ((0, ripgrep_1.isRgAvailable)()) {
-            return usingRipgrep(input);
+            return usingRipgrep(input, repoPath);
         }
         else {
             console.warn("ripgrep is not available, falling back to system grep.");
-            return usingSystemGrep(input);
+            return usingSystemGrep(input, repoPath);
         }
     },
 };

package/dist/tools/str_replace_editor.d.ts

@@ -17,7 +17,7 @@ export declare function cleanupBackupFiles(repoDir: string): number;
  * Our implementation of Claude's built-in text editor tool
  * https://docs.anthropic.com/en/docs/build-with-claude/tool-use/text-editor-tool
  */
-export declare function strReplaceEditorExecutor(input: StrReplaceInput): Promise<ToolResult>;
+export declare function strReplaceEditorExecutor(input: StrReplaceInput, repoPath: string): Promise<ToolResult>;
 export declare const textEditorTools: Tool[];
 export {};
 //# sourceMappingURL=str_replace_editor.d.ts.map

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,CAuMrB;AAiGD,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,EACtB,QAAQ,EAAE,MAAM,GACf,OAAO,CAAC,UAAU,CAAC,CAkNrB;AAuHD,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}

package/dist/tools/str_replace_editor.js

@@ -139,23 +139,25 @@ function escapeRegExp(text) {
  * Our implementation of Claude's built-in text editor tool
  * https://docs.anthropic.com/en/docs/build-with-claude/tool-use/text-editor-tool
  */
-async function strReplaceEditorExecutor(input) {
+async function strReplaceEditorExecutor(input, repoPath) {
+    const repoDir = repoPath;
     const { path: filePath } = input;
+    const absoluteFilePath = path_1.default.join(repoDir, filePath);
     try {
         let content;
         let lines;
         let newContent;
         switch (input.command) {
             case "view":
-                if (!fs_1.default.existsSync(filePath)) {
+                if (!fs_1.default.existsSync(absoluteFilePath)) {
                     return {
                         result: "Error: File not found",
                         isError: true,
                     };
                 }
                 // Handle directory view
-                if (fs_1.default.statSync(filePath).isDirectory()) {
-                    const files = fs_1.default.readdirSync(filePath);
+                if (fs_1.default.statSync(absoluteFilePath).isDirectory()) {
+                    const files = fs_1.default.readdirSync(absoluteFilePath);
                     return {
                         result: files.join("\n"),
                         isError: false,
@@ -164,7 +166,7 @@ async function strReplaceEditorExecutor(input) {
                 else {
                     // Check if file is binary, which is not supported
                     const { isBinary } = await import("istextorbinary");
-                    const binary = isBinary(filePath);
+                    const binary = isBinary(absoluteFilePath);
                     if (binary) {
                         return {
                             result: "Error: File is binary, which is not supported",
@@ -173,7 +175,7 @@ async function strReplaceEditorExecutor(input) {
                     }
                 }
                 // Handle file view
-                content = fs_1.default.readFileSync(filePath, "utf8");
+                content = fs_1.default.readFileSync(absoluteFilePath, "utf8");
                 lines = content.split("\n");
                 if (input.view_range) {
                     const [start, end] = input.view_range;
@@ -194,13 +196,19 @@ async function strReplaceEditorExecutor(input) {
                 if (input.file_text === undefined || input.file_text === null) {
                     throw new Error("file_text is required for create command");
                 }
-                const parentDir = path_1.default.dirname(filePath);
+                if (filePath.endsWith("test.ts")) {
+                    throw new Error("Creating test.ts files is not allowed. Did you mean spec.ts?");
+                }
+                if (filePath.endsWith("spec.ts") && !filePath.startsWith("tests/")) {
+                    throw new Error("Creating spec.ts files is not allowed outside tests/ directory");
+                }
+                const parentDir = path_1.default.dirname(absoluteFilePath);
                 if (parentDir !== "." && !fs_1.default.existsSync(parentDir)) {
                     // Ensure parent directory exists
                     fs_1.default.mkdirSync(parentDir, { recursive: true });
                 }
-                fs_1.default.writeFileSync(filePath, input.file_text);
-                let createTypescriptResult = await (0, web_1.runTypescriptCompiler)();
+                fs_1.default.writeFileSync(absoluteFilePath, input.file_text);
+                let createTypescriptResult = await (0, web_1.runTypescriptCompiler)(repoDir);
                 if (!createTypescriptResult.success) {
                     return {
                         result: `File ${filePath} has been created. However, type checks are failing with errors:\n${createTypescriptResult.errors.join("\n")}`,
@@ -219,8 +227,8 @@ async function strReplaceEditorExecutor(input) {
                     // "" is valid as new_str, so we check for nullish -- not falsy
                     throw new Error("new_str is required for str_replace command");
                 }
-                createBackup(filePath);
-                content = fs_1.default.readFileSync(filePath, "utf8");
+                createBackup(absoluteFilePath);
+                content = fs_1.default.readFileSync(absoluteFilePath, "utf8");
                 // Normalize newlines in both the content and search string
                 const normalizedContent = content.replace(/\r\n/g, "\n");
                 const normalizedOldStr = input.old_str
@@ -252,8 +260,8 @@ async function strReplaceEditorExecutor(input) {
                         };
                     }
                     newContent = normalizedContent.replace(normalizedOldStr, input.new_str);
-                    fs_1.default.writeFileSync(filePath, newContent);
-                    let strReplaceTypescriptResult = await (0, web_1.runTypescriptCompiler)();
+                    fs_1.default.writeFileSync(absoluteFilePath, newContent);
+                    let strReplaceTypescriptResult = await (0, web_1.runTypescriptCompiler)(repoDir);
                     if (!strReplaceTypescriptResult.success) {
                         return {
                             result: `Edits to file ${filePath} have been applied. However, type checks are failing with errors:\n${strReplaceTypescriptResult.errors.join("\n")}`,
@@ -271,8 +279,8 @@ async function strReplaceEditorExecutor(input) {
                 if (input.insert_line === undefined || !input.new_str) {
                     throw new Error("insert_line and new_str are required for insert command");
                 }
-                createBackup(filePath);
-                content = fs_1.default.readFileSync(filePath, "utf8");
+                createBackup(absoluteFilePath);
+                content = fs_1.default.readFileSync(absoluteFilePath, "utf8");
                 lines = content.split("\n");
                 if (input.insert_line < 1) {
                     throw new Error("insert_line must be greater than or equal to 1 (line numbers are 1-indexed).");
@@ -281,8 +289,8 @@ async function strReplaceEditorExecutor(input) {
                     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"));
-                let insertTypescriptResult = await (0, web_1.runTypescriptCompiler)();
+                fs_1.default.writeFileSync(absoluteFilePath, lines.join("\n"));
+                let insertTypescriptResult = await (0, web_1.runTypescriptCompiler)(repoDir);
                 if (!insertTypescriptResult.success) {
                     return {
                         result: `Insertion in file ${filePath} was applied. However, type checks are failing with errors:\n${insertTypescriptResult.errors.join("\n")}`,
@@ -296,8 +304,8 @@ async function strReplaceEditorExecutor(input) {
                     };
                 }
             case "undo_edit":
-                if (hasBackup(filePath)) {
-                    restoreBackup(filePath);
+                if (hasBackup(absoluteFilePath)) {
+                    restoreBackup(absoluteFilePath);
                     return {
                         result: `Successfully restored ${filePath} from backup`,
                         isError: false,
@@ -335,11 +343,11 @@ File contents are returned with line numbers, starting from 1.
             path: zod_1.z.string().describe("The path to the file or directory to view."),
         }),
     },
-    execute: async (input) => {
+    execute: async (input, repoPath) => {
         return strReplaceEditorExecutor({
             command: "view",
             path: input.path,
-        });
+        }, repoPath);
     },
 };
 const fileCreateTool = {
@@ -351,12 +359,12 @@ const fileCreateTool = {
             file_text: zod_1.z.string().describe("The contents of the new file."),
         }),
     },
-    execute: async (input) => {
+    execute: async (input, repoPath) => {
         return strReplaceEditorExecutor({
             command: "create",
             path: input.path,
             file_text: input.file_text,
-        });
+        }, repoPath);
     },
 };
 const stringReplaceTool = {
@@ -370,13 +378,13 @@ in the file. If old_str is not unique, the tool will return an error.`,
             new_str: zod_1.z.string().describe("The string to replace old_str with."),
         }),
     },
-    execute: async (input) => {
+    execute: async (input, repoPath) => {
         return strReplaceEditorExecutor({
             command: "str_replace",
             path: input.path,
             old_str: input.old_str,
             new_str: input.new_str,
-        });
+        }, repoPath);
     },
 };
 const stringInsertTool = {
@@ -389,17 +397,19 @@ const stringInsertTool = {
                 .number()
                 .int()
                 .min(1)
-                .describe("The line number after which to insert the text (1 for beginning of file)."),
+                .describe(`The line number on which to insert the text (1 for beginning of file).
+To insert a string at the beginning of the file, you should use insert_line = 1.
+To insert a string at the end of the file, you should use insert_line = (total lines + 1).`),
             new_str: zod_1.z.string().describe("The string to insert."),
         }),
     },
-    execute: async (input) => {
+    execute: async (input, repoPath) => {
         return strReplaceEditorExecutor({
             command: "insert",
             path: input.path,
             insert_line: input.insert_line,
             new_str: input.new_str,
-        });
+        }, repoPath);
     },
 };
 exports.textEditorTools = [

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;AAsFnD,eAAO,MAAM,4BAA4B,EAAE,IAqF1C,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;AA2FnD,eAAO,MAAM,4BAA4B,EAAE,IA8G1C,CAAC"}

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

@@ -9,6 +9,7 @@ const promises_1 = __importDefault(require("fs/promises"));
 const zod_1 = require("zod");
 const run_1 = require("../agent/browsing/run");
 const utils_1 = require("../agent/browsing/utils");
+const pw_pause_1 = require("../agent/cua/pw-codegen/pw-pause");
 const web_1 = require("../bin/utils/platform/web");
 const scenarios_1 = require("../bin/utils/scenarios");
 const BrowserAgentSchema = zod_1.z.object({
@@ -87,8 +88,8 @@ exports.generateTestWithBrowserAgent = {
         description: BROWSER_AGENT_DESCRIPTION,
         parameters: BrowserAgentSchema,
     },
-    execute: async (input, trace) => {
-        const repoDir = process.cwd();
+    execute: async (input, repoPath, apiKey, trace, collectArtifacts) => {
+        const repoDir = repoPath;
         const { testName, testSuites, fileName, changeToMake, project } = input;
         try {
             const { projects } = await (0, test_run_1.getAllPlaywrightProjects)(repoDir);
@@ -113,6 +114,7 @@ exports.generateTestWithBrowserAgent = {
                 result: `Test block not found for test name: "${testName}" in file: "${fileName}" with describe blocks: "${testSuites.join(", ")}"`,
             };
         }
+        // Prepare the file for the browser agent
         const fileBackup = await promises_1.default.readFile(fileName, "utf-8");
         try {
             await (0, utils_1.replaceTodoWithCreateTest)(fileName);
@@ -125,6 +127,14 @@ exports.generateTestWithBrowserAgent = {
                 result: `Error running tool: ${error}`,
             };
         }
+        try {
+            // Prepare playwright for codegen
+            console.log("[generateTestWithBrowserAgent] Preparing playwright for codegen");
+            await (0, pw_pause_1.preparePlaywrightForCodegen)(repoDir);
+        }
+        catch (err) {
+            console.warn("[generateTestWithBrowserAgent] Error preparing playwright for codegen", err);
+        }
         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)({
@@ -138,9 +148,13 @@ exports.generateTestWithBrowserAgent = {
             repoDir,
             editFileWithGeneratedCode: false,
         });
-        // Undo the TODO -> createTest changes
+        // Cleanup: Undo the TODO -> createTest changes
+        await (0, pw_pause_1.revertToOriginalPwCode)(repoDir);
         await promises_1.default.writeFile(fileName, fileBackup, "utf-8");
-        const { isError, error, actionsSummary } = toolResult;
+        const { isError, error, actionsSummary, artifacts } = toolResult;
+        if (artifacts) {
+            void collectArtifacts?.(artifacts);
+        }
         if (!isError) {
             return {
                 isError,

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

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/test-run-fetcher/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAcnD,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAOnE;AAED,eAAO,MAAM,sBAAsB,EAAE,IAwHpC,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/test-run-fetcher/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAcnD,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAOnE;AAED,eAAO,MAAM,sBAAsB,EAAE,IA6HpC,CAAC"}

package/dist/tools/test-run-fetcher/index.js

@@ -24,7 +24,7 @@ exports.fetchTestRunReportTool = {
         description: "Fetch details about a test run using its URL",
         parameters: TestRunSchema,
     },
-    execute: async (input) => {
+    execute: async (input, repoPath, apiKey) => {
         const { testRunUrl } = input;
         // Remove query parameters if they exist
         const urlWithoutParams = testRunUrl.split("?")[0] || testRunUrl;
@@ -42,6 +42,7 @@ exports.fetchTestRunReportTool = {
         try {
             data = await (0, utils_1.makeDashboardRequest)({
                 path: `/api/test-runs/${runId}?repo_name=${repoName}`,
+                apiKey,
             });
         }
         catch (error) {

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,IAiDzB,CAAC"}
+{"version":3,"file":"test-run.d.ts","sourceRoot":"","sources":["../../src/tools/test-run.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAc,MAAM,wBAAwB,CAAC;AA4C/D,eAAO,MAAM,WAAW,EAAE,IAiEzB,CAAC"}