@empiricalrun/test-gen 0.64.3 → 0.66.0

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

@@ -1 +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,wBAAgB,WAAW,CAAC,YAAY,EAAE,MAAM,EAAE,sBAMjD;AAED,wBAAsB,4BAA4B,CAChD,QAAQ,EAAE,MAAM,EAChB,OAAO,EAAE;IACP,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,iBAiBF"}
+{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/utils.ts"],"names":[],"mappings":"AACA,OAAO,EACL,eAAe,EACf,cAAc,EACd,IAAI,EACL,MAAM,4BAA4B,CAAC;AAEpC,wBAAgB,0BAA0B,CACxC,QAAQ,EAAE,MAAM,EAChB,UAAU,EAAE,MAAM,GACjB,MAAM,CAOR;AAYD,wBAAgB,WAAW,CACzB,SAAS,EAAE,eAAe,EAAE,EAC5B,KAAK,EAAE,IAAI,EAAE,GACZ,MAAM,CAeR;AAED,wBAAsB,4BAA4B,CAChD,QAAQ,EAAE,MAAM,EAChB,OAAO,EAAE,cAAc,iBAiBxB"}

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

@@ -1,14 +1,37 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.appendBranchNameToQueueUrl = appendBranchNameToQueueUrl;
 exports.getQueueUrl = getQueueUrl;
 exports.sendToolRequestToRemoteQueue = sendToolRequestToRemoteQueue;
 const client_sqs_1 = require("@aws-sdk/client-sqs");
-function getQueueUrl(featureFlags) {
-    const useToolExecutionOnEcs = featureFlags.includes("useToolExecutionOnEcs");
-    if (useToolExecutionOnEcs) {
-        return process.env.TOOL_EXECUTION_SQS_URL_ECS;
+function appendBranchNameToQueueUrl(queueUrl, branchName) {
+    const cleanBranchName = branchName.replace(/[/.]/g, "-");
+    const suffix = cleanBranchName.slice(0, 50);
+    const parts = queueUrl.split("/");
+    const queueName = parts.pop();
+    const queueNameWithoutFifo = queueName?.replace(/\.fifo$/, "");
+    return `${parts.join("/")}/${queueNameWithoutFifo}-${suffix}.fifo`;
+}
+function buildQueueUrlsForPreview(baseQueueUrl) {
+    const branch = process.env.VERCEL_GIT_COMMIT_REF;
+    if (!branch) {
+        throw new Error("VERCEL_GIT_COMMIT_REF is not set");
+    }
+    const queueUrl = appendBranchNameToQueueUrl(baseQueueUrl, branch);
+    console.log("queueUrl for preview", queueUrl);
+    return queueUrl;
+}
+function getQueueUrl(toolCalls, tools) {
+    const needsBrowser = toolCalls.some((toolCall) => tools.find((t) => t.schema.name === toolCall.name)?.needsBrowser);
+    const baseQueueUrl = needsBrowser
+        ? process.env.TOOL_EXECUTION_SQS_URL_BROWSER
+        : process.env.TOOL_EXECUTION_SQS_URL;
+    if (process.env.VERCEL_ENV !== "preview") {
+        return baseQueueUrl;
+    }
+    else {
+        return buildQueueUrlsForPreview(baseQueueUrl);
     }
-    return process.env.TOOL_EXECUTION_SQS_URL;
 }
 async function sendToolRequestToRemoteQueue(queueUrl, payload) {
     const sqs = new client_sqs_1.SQSClient({

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

@@ -1,9 +1,3 @@
 import type { Tool } from "@empiricalrun/shared-types";
-interface GitHubRepoInfo {
-    owner: string;
-    repo: string;
-}
-export declare function parseGitHubUrl(url: string): GitHubRepoInfo;
 export declare const createPullRequestTool: Tool;
-export {};
 //# sourceMappingURL=commit-and-create-pr.d.ts.map

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,EAEV,IAAI,EAEL,MAAM,4BAA4B,CAAC;AAOpC,UAAU,cAAc;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;CACd;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,GAAG,cAAc,CAiB1D;AAiCD,eAAO,MAAM,qBAAqB,EAAE,IAkFnC,CAAC"}
+{"version":3,"file":"commit-and-create-pr.d.ts","sourceRoot":"","sources":["../../src/tools/commit-and-create-pr.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAEV,IAAI,EAEL,MAAM,4BAA4B,CAAC;AAoCpC,eAAO,MAAM,qBAAqB,EAAE,IAkFnC,CAAC"}

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

@@ -1,25 +1,10 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.createPullRequestTool = void 0;
-exports.parseGitHubUrl = parseGitHubUrl;
 const child_process_1 = require("child_process");
 const zod_1 = require("zod");
 const git_1 = require("../utils/git");
 const utils_1 = require("./utils");
-function parseGitHubUrl(url) {
-    const githubIndex = url.indexOf("github.com");
-    if (githubIndex === -1) {
-        throw new Error("Invalid GitHub repository URL");
-    }
-    const [owner, repo] = url
-        .substring(githubIndex + "github.com/".length)
-        .replace(".git", "")
-        .split("/");
-    if (!owner || !repo) {
-        throw new Error("Invalid GitHub repository URL format - missing owner or repo");
-    }
-    return { owner, repo };
-}
 const createPullRequestSchema = zod_1.z.object({
     pullRequestTitle: zod_1.z
         .string()
@@ -51,6 +36,7 @@ Don't ask the user for this information, just come up with it yourself.
 `,
         parameters: createPullRequestSchema,
     },
+    needsBrowser: false,
     execute: async ({ input, repoPath, apiKey, }) => {
         try {
             const { pullRequestTitle, pullRequestDescription } = input;
@@ -60,25 +46,22 @@ Don't ask the user for this information, just come up with it yourself.
             })
                 .toString()
                 .trim();
-            const { owner, repo } = parseGitHubUrl(repoUrl);
-            const existingPRs = (await (0, utils_1.callGitHubProxy)({
-                method: "GET",
-                url: `https://api.github.com/repos/${owner}/${repo}/pulls`,
-                body: {
-                    head: `${owner}:${branchName}`,
-                    state: "open",
-                },
-                apiKey,
-            }));
+            const { owner, repo } = (0, utils_1.parseGitHubUrl)(repoUrl);
             (0, child_process_1.execSync)(`git push origin ${branchName} --set-upstream`, {
                 cwd: repoPath,
             });
-            const existingPR = existingPRs?.find((pr) => pr.head.ref === branchName);
+            const existingPR = await (0, utils_1.findExistingPR)({
+                owner,
+                repo,
+                branchName,
+                apiKey,
+            });
             if (existingPR) {
-                await (0, utils_1.callGitHubProxy)({
-                    method: "PATCH",
-                    url: `https://api.github.com/repos/${owner}/${repo}/pulls/${existingPR.number}`,
-                    body: { body: descriptionWithTimestamp(pullRequestDescription) },
+                await (0, utils_1.updatePullRequest)({
+                    owner,
+                    repo,
+                    prNumber: existingPR.number,
+                    body: descriptionWithTimestamp(pullRequestDescription),
                     apiKey,
                 });
                 return {
@@ -86,17 +69,15 @@ Don't ask the user for this information, just come up with it yourself.
                     result: `Committed and pushed changes to existing PR: ${existingPR.html_url}`,
                 };
             }
-            const pr = (await (0, utils_1.callGitHubProxy)({
-                method: "POST",
-                url: `https://api.github.com/repos/${owner}/${repo}/pulls`,
-                body: {
-                    title: pullRequestTitle,
-                    head: branchName,
-                    base: "main",
-                    body: descriptionWithTimestamp(pullRequestDescription),
-                },
+            const pr = await (0, utils_1.createPullRequest)({
+                owner,
+                repo,
+                title: pullRequestTitle,
+                head: branchName,
+                base: "main",
+                body: descriptionWithTimestamp(pullRequestDescription),
                 apiKey,
-            }));
+            });
             return {
                 isError: false,
                 result: `Committed and pushed changes to new PR: ${pr.html_url}`,

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,4BAA4B,CAAC;AAepE,eAAO,MAAM,wBAAwB,EAAE,IAqFtC,CAAC"}
+{"version":3,"file":"diagnosis-fetcher.d.ts","sourceRoot":"","sources":["../../src/tools/diagnosis-fetcher.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAe,MAAM,4BAA4B,CAAC;AAepE,eAAO,MAAM,wBAAwB,EAAE,IAsFtC,CAAC"}

package/dist/tools/diagnosis-fetcher.js

@@ -19,6 +19,7 @@ exports.fetchDiagnosisReportTool = {
         description: "Fetch details about a test case diagnosis using its URL or slug",
         parameters: DiagnosisSchema,
     },
+    needsBrowser: false,
     execute: async (params) => {
         const { input } = params;
         const { repoPath, apiKey } = params;

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,4BAA4B,CAAC;AACpE,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"}
+{"version":3,"file":"download-build.d.ts","sourceRoot":"","sources":["../../src/tools/download-build.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAe,MAAM,4BAA4B,CAAC;AACpE,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,IAmC/B,CAAC"}

package/dist/tools/download-build.js

@@ -14,6 +14,7 @@ have a build URL, you can try getting the environment details with the getEnviro
 Environment details will include the build URL.`,
         parameters: exports.downloadBuildToolSchema,
     },
+    needsBrowser: false,
     execute: async (params) => {
         const { input } = params;
         const { repoPath, apiKey } = params;

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,EAEV,IAAI,EAGL,MAAM,4BAA4B,CAAC;AA8HpC,eAAO,MAAM,QAAQ,EAAE,IA2BtB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/grep/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAEV,IAAI,EAGL,MAAM,4BAA4B,CAAC;AA8HpC,eAAO,MAAM,QAAQ,EAAE,IA4BtB,CAAC"}

package/dist/tools/grep/index.js

@@ -124,6 +124,7 @@ 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,
     },
+    needsBrowser: false,
     execute: async ({ input, repoPath, }) => {
         if ((0, ripgrep_1.isRgAvailable)()) {
             return usingRipgrep(input, repoPath);

package/dist/tools/list-environments.d.ts

@@ -0,0 +1,3 @@
+import type { Tool } from "@empiricalrun/shared-types";
+export declare const listEnvironmentsTool: Tool;
+//# sourceMappingURL=list-environments.d.ts.map

package/dist/tools/list-environments.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"list-environments.d.ts","sourceRoot":"","sources":["../../src/tools/list-environments.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAEV,IAAI,EAEL,MAAM,4BAA4B,CAAC;AAKpC,eAAO,MAAM,oBAAoB,EAAE,IAsDlC,CAAC"}

package/dist/tools/list-environments.js

@@ -0,0 +1,49 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.listEnvironmentsTool = void 0;
+const zod_1 = require("zod");
+const utils_1 = require("./utils");
+exports.listEnvironmentsTool = {
+    schema: {
+        name: "listEnvironments",
+        description: "List all environments and their latest builds for this repository.",
+        parameters: zod_1.z.object({}),
+    },
+    needsBrowser: false,
+    execute: async ({ apiKey, }) => {
+        let response;
+        try {
+            response = await (0, utils_1.makeDashboardRequest)({
+                path: `/api/environments/list`,
+                method: "GET",
+                apiKey,
+            });
+        }
+        catch (error) {
+            return {
+                isError: true,
+                result: `Failed to make API request to fetch environments: ${error instanceof Error ? error.message : String(error)}`,
+            };
+        }
+        try {
+            const data = response;
+            if (data.error) {
+                return {
+                    isError: true,
+                    result: `API error while fetching environments: ${data.error.message}`,
+                    data: data.error,
+                };
+            }
+            return {
+                result: JSON.stringify(data.data),
+                isError: false,
+            };
+        }
+        catch (error) {
+            return {
+                isError: true,
+                result: `Failed to parse or validate environment response data: ${error instanceof Error ? error.message : String(error)}`,
+            };
+        }
+    },
+};

package/dist/tools/str_replace_editor.d.ts

@@ -1,4 +1,4 @@
-import type { Tool, ToolResult } from "@empiricalrun/shared-types";
+import type { CollectArtifacts, Tool, ToolResult } from "@empiricalrun/shared-types";
 interface StrReplaceInput {
     command: string;
     path: string;
@@ -12,9 +12,10 @@ interface StrReplaceInput {
  * 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, repoPath, }: {
+export declare function strReplaceEditorExecutor({ input, repoPath, collectArtifacts, }: {
     input: StrReplaceInput;
     repoPath: string;
+    collectArtifacts?: CollectArtifacts;
 }): Promise<ToolResult>;
 export declare const textEditorTools: Tool[];
 export {};

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,KAAK,EAAE,IAAI,EAAe,UAAU,EAAE,MAAM,4BAA4B,CAAC;AAOhF,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;AA6UD;;;GAGG;AACH,wBAAsB,wBAAwB,CAAC,EAC7C,KAAK,EACL,QAAQ,GACT,EAAE;IACD,KAAK,EAAE,eAAe,CAAC;IACvB,QAAQ,EAAE,MAAM,CAAC;CAClB,GAAG,OAAO,CAAC,UAAU,CAAC,CA2CtB;AA8HD,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":"AAAA,OAAO,KAAK,EAEV,gBAAgB,EAChB,IAAI,EAEJ,UAAU,EACX,MAAM,4BAA4B,CAAC;AAsCpC,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;AAiWD;;;GAGG;AACH,wBAAsB,wBAAwB,CAAC,EAC7C,KAAK,EACL,QAAQ,EACR,gBAAgB,GACjB,EAAE;IACD,KAAK,EAAE,eAAe,CAAC;IACvB,QAAQ,EAAE,MAAM,CAAC;IACjB,gBAAgB,CAAC,EAAE,gBAAgB,CAAC;CACrC,GAAG,OAAO,CAAC,UAAU,CAAC,CA8CtB;AAqID,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}

package/dist/tools/str_replace_editor.js

@@ -9,6 +9,31 @@ const fs_1 = __importDefault(require("fs"));
 const path_1 = __importDefault(require("path"));
 const zod_1 = require("zod");
 const web_1 = require("../bin/utils/platform/web");
+const git_1 = require("../utils/git");
+/**
+ * Helper function to collect git patch artifacts for file modifications
+ */
+async function collectGitPatchArtifact(filePath, repoDir, operation, collectArtifacts) {
+    if (!collectArtifacts)
+        return;
+    try {
+        const gitPatch = operation === "create"
+            ? (0, git_1.getGitDiffForNewFile)(filePath, repoDir)
+            : (0, git_1.getGitDiff)(filePath, repoDir);
+        if (gitPatch.trim()) {
+            const patchArtifact = {
+                name: `${path_1.default.basename(filePath, path_1.default.extname(filePath))}_${operation}.patch`,
+                contentType: "text/plain",
+                data: Buffer.from(gitPatch, "utf-8"),
+            };
+            collectArtifacts([patchArtifact]);
+        }
+    }
+    catch (error) {
+        // Ignore git diff errors, don't fail the operation
+        console.warn(`Failed to generate git patch for ${filePath}:`, error);
+    }
+}
 /**
  * While running str_replace command, we've seen LLM can struggle to send unique old_str.
  * This function tries to find unique contexts for each occurrence of old_str, so that the error
@@ -115,7 +140,7 @@ async function fileViewExecutor({ input, filePath, absoluteFilePath, }) {
         isError: false,
     };
 }
-async function fileCreateExecutor({ input, filePath, absoluteFilePath, repoDir, }) {
+async function fileCreateExecutor({ input, filePath, absoluteFilePath, repoDir, collectArtifacts, }) {
     if (input.file_text === undefined || input.file_text === null) {
         return {
             result: "Error: file_text is required for create command",
@@ -149,6 +174,8 @@ the required directories recursively for the new file.`,
         fs_1.default.mkdirSync(parentDir, { recursive: true });
     }
     fs_1.default.writeFileSync(absoluteFilePath, input.file_text);
+    // Collect git patch artifact
+    await collectGitPatchArtifact(filePath, repoDir, "create", collectArtifacts);
     let createTypescriptResult = await (0, web_1.runTypescriptCompiler)(repoDir);
     if (!createTypescriptResult.success) {
         return {
@@ -163,7 +190,7 @@ the required directories recursively for the new file.`,
         };
     }
 }
-async function fileStrReplaceExecutor({ input, filePath, absoluteFilePath, repoDir, }) {
+async function fileStrReplaceExecutor({ input, filePath, absoluteFilePath, repoDir, collectArtifacts, }) {
     if (!fs_1.default.existsSync(absoluteFilePath)) {
         return {
             result: `Error: File ${filePath} not found. Please provide relative file path to the Repository.`,
@@ -216,6 +243,8 @@ async function fileStrReplaceExecutor({ input, filePath, absoluteFilePath, repoD
         }
         const newContent = normalizedContent.replace(normalizedOldStr, input.new_str);
         fs_1.default.writeFileSync(absoluteFilePath, newContent);
+        // Collect git patch artifact
+        await collectGitPatchArtifact(filePath, repoDir, "replace", collectArtifacts);
         let strReplaceTypescriptResult = await (0, web_1.runTypescriptCompiler)(repoDir);
         if (!strReplaceTypescriptResult.success) {
             return {
@@ -231,7 +260,7 @@ async function fileStrReplaceExecutor({ input, filePath, absoluteFilePath, repoD
         }
     }
 }
-async function fileInsertExecutor({ input, filePath, absoluteFilePath, repoDir, }) {
+async function fileInsertExecutor({ input, filePath, absoluteFilePath, repoDir, collectArtifacts, }) {
     if (!fs_1.default.existsSync(absoluteFilePath)) {
         return {
             result: `Error: File ${filePath} not found. Please provide relative file path to the Repository.`,
@@ -260,6 +289,8 @@ async function fileInsertExecutor({ input, filePath, absoluteFilePath, repoDir,
     }
     lines.splice(input.insert_line - 1, 0, input.new_str);
     fs_1.default.writeFileSync(absoluteFilePath, lines.join("\n"));
+    // Collect git patch artifact
+    await collectGitPatchArtifact(filePath, repoDir, "insert", collectArtifacts);
     let insertTypescriptResult = await (0, web_1.runTypescriptCompiler)(repoDir);
     if (!insertTypescriptResult.success) {
         return {
@@ -278,7 +309,7 @@ async function fileInsertExecutor({ input, filePath, absoluteFilePath, repoDir,
  * 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, repoPath, }) {
+async function strReplaceEditorExecutor({ input, repoPath, collectArtifacts, }) {
     const repoDir = repoPath;
     const { path: filePath } = input;
     const absoluteFilePath = path_1.default.join(repoDir, filePath);
@@ -292,6 +323,7 @@ async function strReplaceEditorExecutor({ input, repoPath, }) {
                     filePath,
                     absoluteFilePath,
                     repoDir,
+                    collectArtifacts,
                 });
             case "str_replace":
                 return fileStrReplaceExecutor({
@@ -299,6 +331,7 @@ async function strReplaceEditorExecutor({ input, repoPath, }) {
                     filePath,
                     absoluteFilePath,
                     repoDir,
+                    collectArtifacts,
                 });
             case "insert":
                 return fileInsertExecutor({
@@ -306,6 +339,7 @@ async function strReplaceEditorExecutor({ input, repoPath, }) {
                     filePath,
                     absoluteFilePath,
                     repoDir,
+                    collectArtifacts,
                 });
             case "undo_edit":
                 return {
@@ -340,6 +374,7 @@ 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."),
         }),
     },
+    needsBrowser: false,
     execute: async (params) => {
         const { input } = params;
         const { repoPath } = params;
@@ -364,9 +399,10 @@ directories "tests", "tests/example", and "tests/example/foo.spec.ts".`,
             file_text: zod_1.z.string().describe("The contents of the new file."),
         }),
     },
+    needsBrowser: false,
     execute: async (params) => {
         const { input } = params;
-        const { repoPath } = params;
+        const { repoPath, collectArtifacts } = params;
         return strReplaceEditorExecutor({
             input: {
                 command: "create",
@@ -374,6 +410,7 @@ directories "tests", "tests/example", and "tests/example/foo.spec.ts".`,
                 file_text: input.file_text,
             },
             repoPath,
+            collectArtifacts,
         });
     },
 };
@@ -388,9 +425,10 @@ 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."),
         }),
     },
+    needsBrowser: false,
     execute: async (params) => {
         const { input } = params;
-        const { repoPath } = params;
+        const { repoPath, collectArtifacts } = params;
         return strReplaceEditorExecutor({
             input: {
                 command: "str_replace",
@@ -399,6 +437,7 @@ in the file. If old_str is not unique, the tool will return an error.`,
                 new_str: input.new_str,
             },
             repoPath,
+            collectArtifacts,
         });
     },
 };
@@ -418,9 +457,10 @@ To insert a string at the end of the file, you should use insert_line = (total l
             new_str: zod_1.z.string().describe("The string to insert."),
         }),
     },
+    needsBrowser: false,
     execute: async (params) => {
         const { input } = params;
-        const { repoPath } = params;
+        const { repoPath, collectArtifacts } = params;
         return strReplaceEditorExecutor({
             input: {
                 command: "insert",
@@ -429,6 +469,7 @@ To insert a string at the end of the file, you should use insert_line = (total l
                 new_str: input.new_str,
             },
             repoPath,
+            collectArtifacts,
         });
     },
 };

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

@@ -1,3 +1,3 @@
-import { Tool } from "@empiricalrun/shared-types";
+import type { Tool } from "@empiricalrun/shared-types";
 export declare const generateTestWithBrowserAgent: Tool;
 //# sourceMappingURL=test-gen-browser.d.ts.map

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,EAAoB,IAAI,EAAE,MAAM,4BAA4B,CAAC;AAmGpE,eAAO,MAAM,4BAA4B,EAAE,IA0I1C,CAAC"}
+{"version":3,"file":"test-gen-browser.d.ts","sourceRoot":"","sources":["../../src/tools/test-gen-browser.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACV,IAAI,EAGL,MAAM,4BAA4B,CAAC;AAoGpC,eAAO,MAAM,4BAA4B,EAAE,IAoM1C,CAAC"}

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

@@ -95,7 +95,8 @@ exports.generateTestWithBrowserAgent = {
         description: BROWSER_AGENT_DESCRIPTION,
         parameters: BrowserAgentSchema,
     },
-    execute: async ({ input, repoPath, trace, collectArtifacts, environmentOverrides = {}, }) => {
+    needsBrowser: true,
+    execute: async ({ input, repoPath, trace, collectArtifacts, environmentOverrides = {}, featureFlags, }) => {
         const { filePath, project } = input;
         const absoluteFilePath = path_1.default.join(repoPath, filePath);
         if (!fs_1.default.existsSync(absoluteFilePath)) {
@@ -159,9 +160,17 @@ exports.generateTestWithBrowserAgent = {
                 BUILD_URL: input.buildUrl,
             };
         }
+        if (featureFlags.includes("headed_codegen_in_browser_agent")) {
+            // Running browser agent in headed implies using page.pause method for codegen
+            // which improves codegen quality when script injection fails (iframes, etc.)
+            envOverrides = {
+                ...envOverrides,
+                RUN_PLAYWRIGHT_HEADED: "true",
+            };
+        }
         const testGenToken = (0, scenarios_1.buildTokenFromOptions)({ name: testName, file: filePath, prompt: input.changeToMake }, { useComputerUseAgent: true });
         console.log("[generateTestWithBrowserAgent] Validations passed, starting agent");
-        const toolResult = await (0, run_1.generateTestsUsingMasterAgent)({
+        const agentResult = await (0, run_1.runBrowsingAgent)({
             testCaseName: testName,
             testCaseSuites: testSuites,
             testFilePath: filePath,
@@ -176,27 +185,64 @@ exports.generateTestWithBrowserAgent = {
         // Cleanup: Undo the TODO -> createTest changes
         await (0, pw_pause_1.revertToOriginalPwCode)(repoPath);
         fs_1.default.writeFileSync(absoluteFilePath, fileBackup, "utf-8");
-        const { isError, error, actionsSummary, artifacts } = toolResult;
+        const { isError, error, result: toolResult, artifacts, usage, } = agentResult;
         if (artifacts) {
             void collectArtifacts?.(artifacts);
         }
-        if (!isError) {
+        if (isError || !toolResult) {
             return {
                 isError,
-                result: `Browser agent has finished running. Here is the summary of actions it took
-and the generated Playwright code:
-
-${actionsSummary}
+                result: `Browser agent failed to run successfully. Here is the error:
+${error}
 `,
             };
         }
+        let result = [
+            {
+                type: "text",
+                text: "Browser agent has finished running. Here is the summary of actions it took, and the generated Playwright code:",
+            },
+            ...toolResult.map((item) => {
+                if (item.type === "text") {
+                    return {
+                        type: "text",
+                        text: item.text,
+                    };
+                }
+                else if (item.type === "screenshot") {
+                    const bufferData = Buffer.isBuffer(item.screenshot.data)
+                        ? item.screenshot.data
+                        : Buffer.from(item.screenshot.data);
+                    return {
+                        type: "image/png",
+                        base64Data: `${bufferData.toString("base64")}`,
+                    };
+                }
+                else {
+                    return {
+                        type: "text",
+                        text: `Unknown item type: ${item}`,
+                    };
+                }
+            }),
+        ];
+        const isMultiModalEnabled = featureFlags.includes("multi_modal_tool_result_from_browser_agent");
+        if (isMultiModalEnabled) {
+            return {
+                isError,
+                result,
+                usage,
+            };
+        }
         else {
+            const textOnlyResult = result
+                .filter((item) => item.type === "text")
+                .map((item) => item.text)
+                .join("\n");
             return {
                 isError,
-                result: `Browser agent failed to run successfully. Here is the error:
-
-${error}
-`,
+                result: textOnlyResult,
+                usage,
             };
         }
     },

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,4BAA4B,CAAC;AAcvD,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAOnE;AAED,eAAO,MAAM,sBAAsB,EAAE,IA+HpC,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,4BAA4B,CAAC;AAcvD,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAOnE;AAED,eAAO,MAAM,sBAAsB,EAAE,IAgIpC,CAAC"}

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

@@ -24,6 +24,7 @@ exports.fetchTestRunReportTool = {
         description: "Fetch details about a test run using its URL",
         parameters: TestRunSchema,
     },
+    needsBrowser: false,
     execute: async ({ input, apiKey, }) => {
         const { testRunUrl } = input;
         // Remove query parameters if they exist

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,EAAc,MAAM,4BAA4B,CAAC;AAwDnE,eAAO,MAAM,WAAW,EAAE,IAoFzB,CAAC"}
+{"version":3,"file":"test-run.d.ts","sourceRoot":"","sources":["../../src/tools/test-run.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAc,MAAM,4BAA4B,CAAC;AAwDnE,eAAO,MAAM,WAAW,EAAE,IAqFzB,CAAC"}

package/dist/tools/test-run.js

@@ -47,6 +47,7 @@ exports.runTestTool = {
         description: "Run a test",
         parameters: RunTestSchema,
     },
+    needsBrowser: true,
     execute: async ({ input, repoPath, collectArtifacts, environmentOverrides = {}, }) => {
         let reportUrl = undefined;
         let envOverrides = { ...environmentOverrides };

package/dist/tools/upgrade-packages/index.d.ts

@@ -0,0 +1,3 @@
+import type { Tool } from "@empiricalrun/shared-types";
+export declare const upgradePackagesTool: Tool;
+//# sourceMappingURL=index.d.ts.map

package/dist/tools/upgrade-packages/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/upgrade-packages/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,4BAA4B,CAAC;AA+BvD,eAAO,MAAM,mBAAmB,EAAE,IA+GjC,CAAC"}