@empiricalrun/test-gen 0.63.0 → 0.64.0

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # @empiricalrun/test-gen
 
+## 0.64.0
+
+### Minor Changes
+
+- 1a5ec8d: feat: stateless tool execution and named args for tool execute fn
+
+### Patch Changes
+
+- e5b9f7e: feat: consume repoPath in browser agent
+- 43dc453: fix: record locators before execution for more reliable codegen
+- 24d9415: fix: error message in browser agent tool call + removed unused code
+- Updated dependencies [1a5ec8d]
+  - @empiricalrun/llm@0.17.2
+
 ## 0.63.0
 
 ### Minor Changes

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

@@ -1 +1 @@
-{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAc3D,KAAK,iBAAiB,GAAG;IACvB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,yBAAyB,EAAE,OAAO,CAAC;CACpC,CAAC;AAEF,wBAAsB,8BAA8B,CAAC,EACnD,gBAAgB,EAChB,OAAO,EACP,YAAY,GACb,EAAE;IACD,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;CACtB,GAAG,OAAO,CAAC,MAAM,CAAC,CAQlB;AAED,wBAAsB,6BAA6B,CAAC,EAClD,YAAY,EACZ,cAAc,EACd,YAAY,EACZ,gBAAgB,EAChB,WAAW,EACX,YAAY,EACZ,OAAO,EACP,OAAO,EACP,yBAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC;IAC7B,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,CAAC,EAAE,aAAa,EAAE,CAAC;CAC7B,CAAC,CAmFD"}
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAe3D,KAAK,iBAAiB,GAAG;IACvB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,yBAAyB,EAAE,OAAO,CAAC;CACpC,CAAC;AAEF,wBAAsB,8BAA8B,CAAC,EACnD,gBAAgB,EAChB,OAAO,EACP,YAAY,GACb,EAAE;IACD,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;CACtB,GAAG,OAAO,CAAC,MAAM,CAAC,CAQlB;AAED,wBAAsB,6BAA6B,CAAC,EAClD,YAAY,EACZ,cAAc,EACd,YAAY,EACZ,gBAAgB,EAChB,WAAW,EACX,YAAY,EACZ,OAAO,EACP,OAAO,EACP,yBAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC;IAC7B,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,CAAC,EAAE,aAAa,EAAE,CAAC;CAC7B,CAAC,CA2FD"}

package/dist/agent/browsing/run.js

@@ -8,6 +8,7 @@ exports.generateTestsUsingMasterAgent = generateTestsUsingMasterAgent;
 const test_run_1 = require("@empiricalrun/test-run");
 const detect_port_1 = __importDefault(require("detect-port"));
 const fs_1 = __importDefault(require("fs"));
+const path_1 = __importDefault(require("path"));
 const web_1 = require("../../bin/utils/platform/web");
 const server_1 = require("../../file/server");
 const exec_1 = require("../../utils/exec");
@@ -18,8 +19,10 @@ async function convertProjectsFilterToProject({ pwProjectsFilter, repoDir, testF
     return project;
 }
 async function generateTestsUsingMasterAgent({ testCaseName, testCaseSuites, testFilePath, filePathToUpdate, projectName, testGenToken, repoDir, traceId, editFileWithGeneratedCode, }) {
-    if (!fs_1.default.existsSync(testFilePath)) {
-        throw new Error(`File for master agent to run not found: ${testFilePath}`);
+    const absTestFilePath = path_1.default.join(repoDir, testFilePath);
+    if (!fs_1.default.existsSync(absTestFilePath)) {
+        const errorMsg = `File for master agent to run not found: ${testFilePath}`;
+        throw new Error(errorMsg);
     }
     const pm = new exec_1.ProcessManager();
     const availablePort = await (0, detect_port_1.default)(3030);
@@ -32,7 +35,8 @@ async function generateTestsUsingMasterAgent({ testCaseName, testCaseSuites, tes
     });
     await fileServer.startFileService();
     fileServer.setFilePath(filePathToUpdate);
-    const pageVar = await (0, web_1.getPageVariableNameFromCreateTest)(filePathToUpdate);
+    const absFilePathToUpdate = path_1.default.join(repoDir, filePathToUpdate);
+    const pageVar = await (0, web_1.getPageVariableNameFromCreateTest)(absFilePathToUpdate);
     const testsDir = "tests";
     const cmd = await (0, test_run_1.runSpecificTestsCmd)({
         tests: [
@@ -66,11 +70,12 @@ async function generateTestsUsingMasterAgent({ testCaseName, testCaseSuites, tes
                 // Enable this for the page.pause approach in codegen
                 PW_CODEGEN_NO_INSPECTOR: "1",
             },
+            cwd: repoDir,
         });
     }
     catch (e) {
         error = `Failed to generate test using master agent ${e}`;
-        console.error(error);
+        console.error(`[generateTestsUsingMasterAgent] ${error}`);
     }
     if (error) {
         // Clean up the file if there is any error
@@ -81,7 +86,7 @@ async function generateTestsUsingMasterAgent({ testCaseName, testCaseSuites, tes
             await (0, web_1.lintErrors)(filePathToUpdate);
         }
         catch (e) {
-            console.error("Failed to remove extra scripts from files post test gen error", e);
+            console.error(`[generateTestsUsingMasterAgent] Failed to remove extra scripts from files post test gen error:`, e);
         }
     }
     await fileServer.stop();

package/dist/agent/browsing/utils.d.ts

@@ -4,7 +4,7 @@ import { Page } from "playwright";
 import { PlaywrightTestConfig } from "playwright/test";
 export declare function isRegExp(obj: any): obj is RegExp;
 export declare function prepareBrowsingAgentTask(steps: string[]): string;
-export declare function replaceTodoWithCreateTest(testFilePath: string): Promise<void>;
+export declare function replaceTodoWithCreateTest(testFilePath: string, repoDir: string): Promise<void>;
 export declare function markTestAsOnly({ testCaseName, testCaseSuites, specPath, }: {
     testCaseName: string;
     testCaseSuites: string[];

package/dist/agent/browsing/utils.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,4BAA4B,CAAC;AAI3D,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAClC,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAsBvD,wBAAgB,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,IAAI,MAAM,CAKhD;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,EAAE,UAIvD;AAiFD,wBAAsB,yBAAyB,CAAC,YAAY,EAAE,MAAM,iBAsBnE;AAED,wBAAsB,cAAc,CAAC,EACnC,YAAY,EACZ,cAAc,EACd,QAAQ,GACT,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,QAAQ,EAAE,MAAM,CAAC;CAClB,iBAoBA;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,QAAQ,EACR,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,MAAM,CAAC,CAyDlB;AAyBD,wBAAsB,wBAAwB,CAAC,IAAI,EAAE,IAAI,iBA8HxD;AAED;;;GAGG;AACH,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,oBAAoB,CAAC,CA2B/B;AAWD,wBAAsB,oBAAoB,CACxC,gBAAgB,EAAE,oBAAoB,GACrC,OAAO,CAAC,MAAM,EAAE,CAAC,CAQnB;AAED;;;;;GAKG;AACH,wBAAsB,iBAAiB,CACrC,YAAY,EAAE,MAAM,EACpB,gBAAgB,EAAE,oBAAoB,EACtC,gBAAgB,GAAE,MAAM,EAAU,GACjC,OAAO,CAAC,MAAM,CAAC,CA+CjB"}
+{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,4BAA4B,CAAC;AAI3D,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAClC,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAsBvD,wBAAgB,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,IAAI,MAAM,CAKhD;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,EAAE,UAIvD;AAiFD,wBAAsB,yBAAyB,CAC7C,YAAY,EAAE,MAAM,EACpB,OAAO,EAAE,MAAM,iBAwBhB;AAED,wBAAsB,cAAc,CAAC,EACnC,YAAY,EACZ,cAAc,EACd,QAAQ,GACT,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,QAAQ,EAAE,MAAM,CAAC;CAClB,iBAoBA;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,QAAQ,EACR,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,MAAM,CAAC,CAyDlB;AAyBD,wBAAsB,wBAAwB,CAAC,IAAI,EAAE,IAAI,iBA8HxD;AAED;;;GAGG;AACH,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,oBAAoB,CAAC,CA2B/B;AAWD,wBAAsB,oBAAoB,CACxC,gBAAgB,EAAE,oBAAoB,GACrC,OAAO,CAAC,MAAM,EAAE,CAAC,CAQnB;AAED;;;;;GAKG;AACH,wBAAsB,iBAAiB,CACrC,YAAY,EAAE,MAAM,EACpB,gBAAgB,EAAE,oBAAoB,EACtC,gBAAgB,GAAE,MAAM,EAAU,GACjC,OAAO,CAAC,MAAM,CAAC,CA+CjB"}

package/dist/agent/browsing/utils.js

@@ -93,10 +93,11 @@ async function prepareFileForUpdateScenario({ testCase, specPath, trace, }) {
     });
     return createTestFilePath;
 }
-async function replaceTodoWithCreateTest(testFilePath) {
+async function replaceTodoWithCreateTest(testFilePath, repoDir) {
     // This method is an alternative to prepareFileForUpdateScenario
     // TODO: Does not support scoped variables and updates in POM files
-    const fileContent = fs_1.default.readFileSync(testFilePath, "utf-8");
+    const absoluteTestFilePath = path_1.default.join(repoDir, testFilePath);
+    const fileContent = fs_1.default.readFileSync(absoluteTestFilePath, "utf-8");
     const todoRegex = /\/\/ TODO\(agent(?:\s+on\s+(\w+))?\):\s*(.*)/;
     const todoMatch = fileContent.match(todoRegex);
     if (!todoMatch) {
@@ -104,8 +105,8 @@ async function replaceTodoWithCreateTest(testFilePath) {
     }
     const [, pageVarName] = todoMatch;
     const pageVariable = pageVarName || "page"; // Default to "page" if not specified
-    fs_1.default.writeFileSync(testFilePath, fileContent.replace(todoRegex, (_, __, todoText) => `await createTest("${todoText.replace(/"/g, '\\"')}", ${pageVariable});`));
-    await addImportForCreateTest(testFilePath);
+    fs_1.default.writeFileSync(absoluteTestFilePath, fileContent.replace(todoRegex, (_, __, todoText) => `await createTest("${todoText.replace(/"/g, '\\"')}", ${pageVariable});`));
+    await addImportForCreateTest(absoluteTestFilePath);
 }
 async function markTestAsOnly({ testCaseName, testCaseSuites, specPath, }) {
     const testFileContent = fs_1.default.readFileSync(specPath, "utf-8");

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

@@ -1 +1 @@
-{"version":3,"file":"computer.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/computer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,wBAAwB,EAAE,MAAM,0CAA0C,CAAC;AACpF,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAEvC,OAAO,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAE3D,KAAK,oBAAoB,GACrB,wBAAwB,CAAC,KAAK,GAC9B,wBAAwB,CAAC,WAAW,GACpC,wBAAwB,CAAC,IAAI,GAC7B,wBAAwB,CAAC,QAAQ,GACjC,wBAAwB,CAAC,IAAI,GAC7B,wBAAwB,CAAC,UAAU,GACnC,wBAAwB,CAAC,MAAM,GAC/B,wBAAwB,CAAC,IAAI,GAC7B,wBAAwB,CAAC,IAAI,CAAC;AAElC,wBAAsB,aAAa,CAAC,IAAI,EAAE,IAAI,mBAG7C;AAgCD,wBAAsB,kBAAkB,CACtC,IAAI,EAAE,IAAI,EACV,MAAM,EAAE,oBAAoB,GAAG;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,GAAG,EAAE,MAAM,CAAA;CAAE,EAC5D,OAAO,EAAE,qBAAqB,GAC7B,OAAO,CAAC;IACT,aAAa,EAAE,MAAM,CAAC;IACtB,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC,CAkJD"}
+{"version":3,"file":"computer.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/computer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,wBAAwB,EAAE,MAAM,0CAA0C,CAAC;AACpF,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAEvC,OAAO,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAE3D,KAAK,oBAAoB,GACrB,wBAAwB,CAAC,KAAK,GAC9B,wBAAwB,CAAC,WAAW,GACpC,wBAAwB,CAAC,IAAI,GAC7B,wBAAwB,CAAC,QAAQ,GACjC,wBAAwB,CAAC,IAAI,GAC7B,wBAAwB,CAAC,UAAU,GACnC,wBAAwB,CAAC,MAAM,GAC/B,wBAAwB,CAAC,IAAI,GAC7B,wBAAwB,CAAC,IAAI,CAAC;AAElC,wBAAsB,aAAa,CAAC,IAAI,EAAE,IAAI,mBAG7C;AAgCD,wBAAsB,kBAAkB,CACtC,IAAI,EAAE,IAAI,EACV,MAAM,EAAE,oBAAoB,GAAG;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,GAAG,EAAE,MAAM,CAAA;CAAE,EAC5D,OAAO,EAAE,qBAAqB,GAC7B,OAAO,CAAC;IACT,aAAa,EAAE,MAAM,CAAC;IACtB,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC,CAmKD"}

package/dist/agent/cua/computer.js

@@ -40,14 +40,21 @@ async function executeModelAction(page, action, codegen) {
     let actionCode = "";
     let actionSummary = "";
     try {
+        // For every action, this method:
+        // 1. Record the action in codegen
+        //    -> This allows codegen to find locators for the element
+        //       before the action is executed - execution can be a destructive
+        //       operation that modifies the DOM
+        // 2. Execute the action
+        // 3. Add to action summary
+        // 4. Add to action code
         switch (actionType) {
-            // For every action, this method 1. executes the action,
-            // 2. adds to action summary, and 3. adds to action code
             case "goto": {
                 const { url } = action;
+                await codegen.recordAction({ type: "goto", url });
                 await page.goto(url);
                 actionSummary = `Navigated page to ${url}`;
-                actionCode = await codegen.codeForAction({ type: "goto", url });
+                actionCode = await codegen.getCodeForLastAction();
                 break;
             }
             case "click": {
@@ -61,20 +68,23 @@ async function executeModelAction(page, action, codegen) {
                     pwButton = "middle";
                 }
                 if (pwButton) {
+                    await codegen.recordAction({ type: "click", x, y });
                     await page.mouse.click(x, y, { button: pwButton });
-                    actionCode = await codegen.codeForAction({ type: "click", x, y });
                     actionSummary = `Click at (${x}, ${y}) with button '${button}'`;
+                    actionCode = await codegen.getCodeForLastAction();
                 }
                 if (button === "back" || button === "forward") {
                     if (button === "back") {
+                        await codegen.recordAction({ type: "back" });
                         await page.goBack();
-                        actionCode = await codegen.codeForAction({ type: "back" });
                         actionSummary = `Navigated page back`;
+                        actionCode = await codegen.getCodeForLastAction();
                     }
                     else if (button === "forward") {
+                        await codegen.recordAction({ type: "forward" });
                         await page.goForward();
-                        actionCode = await codegen.codeForAction({ type: "forward" });
                         actionSummary = `Navigated page forward`;
+                        actionCode = await codegen.getCodeForLastAction();
                     }
                 }
                 break;
@@ -82,17 +92,28 @@ async function executeModelAction(page, action, codegen) {
             case "double_click": {
                 const { x, y } = action;
                 console.log(`Action: doubleclick at (${x}, ${y})`);
+                await codegen.recordAction({ type: "doubleclick", x, y });
                 await page.mouse.dblclick(x, y, { button: "left" });
                 actionSummary = `Double click at (${x}, ${y})`;
-                actionCode = await codegen.codeForAction({ type: "doubleclick", x, y });
+                actionCode = await codegen.getCodeForLastAction();
+                break;
+            }
+            case "type": {
+                const { text } = action;
+                console.log(`Action: type text '${text}'`);
+                await codegen.recordAction({ type: "type", text });
+                await page.keyboard.type(text);
+                actionSummary = `Type text '${text}'`;
+                actionCode = await codegen.getCodeForLastAction();
                 break;
             }
             case "move": {
                 const { x, y } = action;
                 console.log(`Action: mouse move to (${x}, ${y})`);
+                await codegen.recordAction({ type: "move", x, y });
                 await page.mouse.move(x, y);
                 actionSummary = `Mouse move to (${x}, ${y})`;
-                actionCode = await codegen.codeForAction({ type: "move", x, y });
+                actionCode = await codegen.getCodeForLastAction();
                 break;
             }
             case "drag": {
@@ -102,32 +123,35 @@ async function executeModelAction(page, action, codegen) {
                 if (!path || path.length === 0) {
                     break;
                 }
+                await codegen.recordAction({ type: "drag", path });
                 await page.mouse.move(path[0].x, path[0].y);
                 await page.mouse.down();
                 for (let i = 1; i < path.length; i++) {
                     await page.mouse.move(path[i].x, path[i].y);
                 }
                 await page.mouse.up();
-                actionCode = await codegen.codeForAction({ type: "drag", path });
+                actionCode = await codegen.getCodeForLastAction();
                 break;
             }
             case "scroll": {
                 const { x, y, scroll_x, scroll_y } = action;
                 console.log(`Action: scroll at (${x}, ${y}) with offsets (scroll_x=${scroll_x}, scroll_y=${scroll_y})`);
-                await page.mouse.move(x, y);
-                await page.evaluate(`window.scrollBy(${scroll_x}, ${scroll_y})`);
-                actionSummary = `Scroll at (${x}, ${y}) with offsets (scroll_x=${scroll_x}, scroll_y=${scroll_y})`;
-                actionCode = await codegen.codeForAction({
+                await codegen.recordAction({
                     type: "scroll",
                     x,
                     y,
                     scroll_x,
                     scroll_y,
                 });
+                await page.mouse.move(x, y);
+                await page.evaluate(`window.scrollBy(${scroll_x}, ${scroll_y})`);
+                actionSummary = `Scroll at (${x}, ${y}) with offsets (scroll_x=${scroll_x}, scroll_y=${scroll_y})`;
+                actionCode = await codegen.getCodeForLastAction();
                 break;
             }
             case "keypress": {
                 const { keys } = action;
+                await codegen.recordAction({ type: "keypress", keys });
                 const mappedKeys = keys.map((k) => {
                     return CUA_KEY_TO_PLAYWRIGHT_KEY[k.toLowerCase()] || k;
                 });
@@ -135,22 +159,15 @@ async function executeModelAction(page, action, codegen) {
                 console.log(`Action: keypress for keys ${keys} -> '${mappedKey}'`);
                 await page.keyboard.press(mappedKey);
                 actionSummary = `Keypress for keys ${keys} (mapped to '${mappedKey}' for Playwright)`;
-                actionCode = await codegen.codeForAction({ type: "keypress", keys });
-                break;
-            }
-            case "type": {
-                const { text } = action;
-                console.log(`Action: type text '${text}'`);
-                await page.keyboard.type(text);
-                actionSummary = `Type text '${text}'`;
-                actionCode = await codegen.codeForAction({ type: "type", text });
+                actionCode = await codegen.getCodeForLastAction();
                 break;
             }
             case "wait": {
                 console.log(`Action: wait`);
+                await codegen.recordAction({ type: "wait" });
                 await page.waitForTimeout(2000);
                 actionSummary = `Wait for 2 seconds`;
-                actionCode = await codegen.codeForAction({ type: "wait" });
+                actionCode = await codegen.getCodeForLastAction();
                 break;
             }
             case "screenshot": {

package/dist/agent/cua/pw-codegen/element-from-point.d.ts

@@ -2,7 +2,9 @@ import type { Page } from "playwright";
 import { Action, BasePlaywrightCodegen } from "./types";
 export declare class ElementFromPointCodegen implements BasePlaywrightCodegen {
     private page;
+    private codeForLastAction;
     initialize(page: Page): Promise<void>;
-    codeForAction(action: Action): Promise<string>;
+    recordAction(action: Action): Promise<void>;
+    getCodeForLastAction(): Promise<string>;
 }
 //# sourceMappingURL=element-from-point.d.ts.map

package/dist/agent/cua/pw-codegen/element-from-point.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"element-from-point.d.ts","sourceRoot":"","sources":["../../../../src/agent/cua/pw-codegen/element-from-point.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAGvC,OAAO,EAAE,MAAM,EAAE,qBAAqB,EAAE,MAAM,SAAS,CAAC;AA4DxD,qBAAa,uBAAwB,YAAW,qBAAqB;IACnE,OAAO,CAAC,IAAI,CAAmB;IACzB,UAAU,CAAC,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC;IAKrC,aAAa,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;CA0DrD"}
+{"version":3,"file":"element-from-point.d.ts","sourceRoot":"","sources":["../../../../src/agent/cua/pw-codegen/element-from-point.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAGvC,OAAO,EAAE,MAAM,EAAE,qBAAqB,EAAE,MAAM,SAAS,CAAC;AA4DxD,qBAAa,uBAAwB,YAAW,qBAAqB;IACnE,OAAO,CAAC,IAAI,CAAmB;IAC/B,OAAO,CAAC,iBAAiB,CAAqB;IAExC,UAAU,CAAC,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC;IAKrC,YAAY,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IA4D3C,oBAAoB,IAAI,OAAO,CAAC,MAAM,CAAC;CAG9C"}

package/dist/agent/cua/pw-codegen/element-from-point.js

@@ -51,35 +51,38 @@ async function getLocatorForFill(page) {
 }
 class ElementFromPointCodegen {
     page;
+    codeForLastAction;
     async initialize(page) {
         this.page = page;
         await (0, utils_1.injectPwLocatorGenerator)(page);
     }
-    async codeForAction(action) {
+    async recordAction(action) {
         try {
             if (action.type === "goto") {
-                return `await page.goto("${action.url}");\n`;
+                this.codeForLastAction = `await page.goto("${action.url}");\n`;
             }
             if (action.type === "click") {
                 const { x, y } = action;
                 if (!this.page) {
                     throw new Error("Page is not initialized");
                 }
+                // TODO: Need execption handling
                 const locator = await getLocatorForClick(this.page, { x, y });
-                return `await page.${locator}.click();\n`;
+                this.codeForLastAction = `await page.${locator}.click();\n`;
             }
             if (action.type === "type") {
                 if (!this.page) {
                     throw new Error("Page is not initialized");
                 }
+                // TODO: Need execption handling
                 const locator = await getLocatorForFill(this.page);
-                return `await page.${locator}.fill("${action.text}");\n`;
+                this.codeForLastAction = `await page.${locator}.fill("${action.text}");\n`;
             }
             if (action.type === "back") {
-                return `await page.goBack();\n`;
+                this.codeForLastAction = `await page.goBack();\n`;
             }
             if (action.type === "forward") {
-                return `await page.goForward();\n`;
+                this.codeForLastAction = `await page.goForward();\n`;
             }
             if (action.type === "doubleclick") {
                 const { x, y } = action;
@@ -87,32 +90,34 @@ class ElementFromPointCodegen {
                     throw new Error("Page is not initialized");
                 }
                 const locator = await getLocatorForClick(this.page, { x, y });
-                return `await page.${locator}.dblclick();\n`;
+                this.codeForLastAction = `await page.${locator}.dblclick();\n`;
             }
             if (action.type === "move") {
                 const { x, y } = action;
-                return `await page.mouse.move(${x}, ${y});\n`;
+                this.codeForLastAction = `await page.mouse.move(${x}, ${y});\n`;
             }
             if (action.type === "drag") {
                 // const { path } = action;
                 // TODO: This needs to be fixed
-                return `No code generated: drag action not supported by code generation`;
+                this.codeForLastAction = `No code generated: drag action not supported by code generation`;
             }
             if (action.type === "scroll") {
-                return `No code to generate: will rely on Playwright's ability to auto-wait`;
+                this.codeForLastAction = `No code to generate: will rely on Playwright's ability to auto-wait`;
             }
             if (action.type === "keypress") {
                 const { keys } = action;
-                return `await page.keyboard.press("${keys.join("+")}");\n`;
+                this.codeForLastAction = `await page.keyboard.press("${keys.join("+")}");\n`;
             }
             if (action.type === "wait") {
-                return `No code to generate: will rely on Playwright's ability to auto-wait`;
+                this.codeForLastAction = `No code to generate: will rely on Playwright's ability to auto-wait`;
             }
-            throw new Error(`Unsupported action: ${JSON.stringify(action)}`);
         }
         catch (e) {
-            return `Error generating code for action: ${e.message || e.toString()}`;
+            this.codeForLastAction = `Error generating code for action: ${e.message || e.toString()}`;
         }
     }
+    async getCodeForLastAction() {
+        return this.codeForLastAction || "No code generated for the last action";
+    }
 }
 exports.ElementFromPointCodegen = ElementFromPointCodegen;

package/dist/agent/cua/pw-codegen/types.d.ts

@@ -41,6 +41,7 @@ export type Action = {
 };
 export interface BasePlaywrightCodegen {
     initialize(page: Page): Promise<void>;
-    codeForAction(action: Action): Promise<string>;
+    recordAction(action: Action): Promise<void>;
+    getCodeForLastAction(): Promise<string>;
 }
 //# sourceMappingURL=types.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../../src/agent/cua/pw-codegen/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAEvC,MAAM,MAAM,MAAM,GACd;IACE,IAAI,EAAE,OAAO,CAAC;IACd,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;CACX,GACD;IACE,IAAI,EAAE,aAAa,CAAC;IACpB,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;CACX,GACD;IACE,IAAI,EAAE,MAAM,CAAC;CACd,GACD;IACE,IAAI,EAAE,SAAS,CAAC;CACjB,GACD;IACE,IAAI,EAAE,MAAM,CAAC;IACb,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;CACX,GACD;IACE,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE;QAAE,CAAC,EAAE,MAAM,CAAC;QAAC,CAAC,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;CAClC,GACD;IACE,IAAI,EAAE,QAAQ,CAAC;IACf,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;IACV,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;CAClB,GACD;IACE,IAAI,EAAE,UAAU,CAAC;IACjB,IAAI,EAAE,MAAM,EAAE,CAAC;CAChB,GACD;IACE,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd,GACD;IACE,IAAI,EAAE,MAAM,CAAC;IACb,GAAG,EAAE,MAAM,CAAC;CACb,GACD;IACE,IAAI,EAAE,MAAM,CAAC;CACd,CAAC;AAEN,MAAM,WAAW,qBAAqB;IACpC,UAAU,CAAC,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACtC,aAAa,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;CAChD"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../../src/agent/cua/pw-codegen/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAEvC,MAAM,MAAM,MAAM,GACd;IAEE,IAAI,EAAE,OAAO,CAAC;IACd,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;CACX,GACD;IACE,IAAI,EAAE,aAAa,CAAC;IACpB,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;CACX,GACD;IACE,IAAI,EAAE,MAAM,CAAC;CACd,GACD;IACE,IAAI,EAAE,SAAS,CAAC;CACjB,GACD;IACE,IAAI,EAAE,MAAM,CAAC;IACb,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;CACX,GACD;IACE,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE;QAAE,CAAC,EAAE,MAAM,CAAC;QAAC,CAAC,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;CAClC,GACD;IACE,IAAI,EAAE,QAAQ,CAAC;IACf,CAAC,EAAE,MAAM,CAAC;IACV,CAAC,EAAE,MAAM,CAAC;IACV,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;CAClB,GACD;IACE,IAAI,EAAE,UAAU,CAAC;IACjB,IAAI,EAAE,MAAM,EAAE,CAAC;CAChB,GACD;IACE,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd,GACD;IACE,IAAI,EAAE,MAAM,CAAC;IACb,GAAG,EAAE,MAAM,CAAC;CACb,GACD;IACE,IAAI,EAAE,MAAM,CAAC;CACd,CAAC;AAEN,MAAM,WAAW,qBAAqB;IACpC,UAAU,CAAC,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACtC,YAAY,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC5C,oBAAoB,IAAI,OAAO,CAAC,MAAM,CAAC,CAAC;CACzC"}

package/dist/file/server.js

@@ -31,7 +31,7 @@ class FileServiceServer {
         return this.artifactsInputs;
     }
     setFilePath(filePath) {
-        this.filePath = filePath;
+        this.filePath = path_1.default.resolve(this.repoDir, filePath);
     }
     async startFileService() {
         const app = (0, express_1.default)();

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

@@ -101,7 +101,13 @@ class ToolCallService {
                 throw new Error("API key is required for tool execution");
             }
             try {
-                const result = await toolExecutor(toolCall.input, this.repoPath, this.apiKey, this.trace, collectArtifactsFn);
+                const result = await toolExecutor({
+                    input: toolCall.input,
+                    repoPath: this.repoPath,
+                    apiKey: this.apiKey,
+                    trace: this.trace,
+                    collectArtifacts: collectArtifactsFn,
+                });
                 const artifacts = await uploadArtifactsQueue.waitForCompletion();
                 toolResults.push({ ...result, artifacts });
                 span?.end({ output: result });

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,IAsFnC,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,IA0FnC,CAAC"}

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

@@ -55,7 +55,7 @@ Don't ask the user for this information, just come up with it yourself.
 `,
         parameters: createPullRequestSchema,
     },
-    execute: async (input, repoPath, apiKey) => {
+    execute: async ({ input, repoPath, apiKey, }) => {
         try {
             const { pullRequestTitle, pullRequestDescription } = input;
             const branchName = await (0, git_1.getCurrentBranchName)(repoPath);

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,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,wBAAwB,CAAC;AAenD,eAAO,MAAM,wBAAwB,EAAE,IAyFtC,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, repoPath, apiKey) => {
+    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();

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,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,IAsC/B,CAAC"}

package/dist/tools/download-build.js

@@ -14,7 +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,
     },
-    execute: async (input, repoPath, apiKey) => {
+    execute: async ({ input, repoPath, apiKey, }) => {
         if (!(await (0, test_build_1.hasDownloadScript)(repoPath))) {
             return {
                 isError: true,

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;AA6DnD,eAAO,MAAM,kBAAkB,EAAE,IA6EhC,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,IAiFhC,CAAC;AAGF,eAAO,MAAM,gBAAgB,EAAE,IAAI,EAAyB,CAAC"}

package/dist/tools/environment-crud.js

@@ -31,7 +31,7 @@ exports.getEnvironmentTool = {
         description: "Fetch details of an existing environment",
         parameters: GetEnvironmentSchema,
     },
-    execute: async (input, repoPath, apiKey) => {
+    execute: async ({ input, repoPath, apiKey, }) => {
         // Get project repo name
         let projectRepoName;
         try {

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;AA6H/D,eAAO,MAAM,QAAQ,EAAE,IAoBtB,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,IAuBtB,CAAC"}

package/dist/tools/grep/index.js

@@ -123,7 +123,7 @@ 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,
     },
-    execute: async (input, repoPath) => {
+    execute: async ({ input, repoPath, }) => {
         if ((0, ripgrep_1.isRgAvailable)()) {
             return usingRipgrep(input, repoPath);
         }

package/dist/tools/str_replace_editor.d.ts

@@ -17,7 +17,10 @@ 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, repoPath: string): Promise<ToolResult>;
+export declare function strReplaceEditorExecutor({ input, repoPath, }: {
+    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,EACtB,QAAQ,EAAE,MAAM,GACf,OAAO,CAAC,UAAU,CAAC,CAkNrB;AAuHD,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,CAAC,EAC7C,KAAK,EACL,QAAQ,GACT,EAAE;IACD,KAAK,EAAE,eAAe,CAAC;IACvB,QAAQ,EAAE,MAAM,CAAC;CAClB,GAAG,OAAO,CAAC,UAAU,CAAC,CAkNtB;AAmID,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}

package/dist/tools/str_replace_editor.js

@@ -139,7 +139,7 @@ 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, repoPath) {
+async function strReplaceEditorExecutor({ input, repoPath, }) {
     const repoDir = repoPath;
     const { path: filePath } = input;
     const absoluteFilePath = path_1.default.join(repoDir, filePath);
@@ -343,11 +343,14 @@ 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, repoPath) => {
+    execute: async ({ input, repoPath, }) => {
         return strReplaceEditorExecutor({
-            command: "view",
-            path: input.path,
-        }, repoPath);
+            input: {
+                command: "view",
+                path: input.path,
+            },
+            repoPath,
+        });
     },
 };
 const fileCreateTool = {
@@ -359,12 +362,15 @@ const fileCreateTool = {
             file_text: zod_1.z.string().describe("The contents of the new file."),
         }),
     },
-    execute: async (input, repoPath) => {
+    execute: async ({ input, repoPath, }) => {
         return strReplaceEditorExecutor({
-            command: "create",
-            path: input.path,
-            file_text: input.file_text,
-        }, repoPath);
+            input: {
+                command: "create",
+                path: input.path,
+                file_text: input.file_text,
+            },
+            repoPath,
+        });
     },
 };
 const stringReplaceTool = {
@@ -378,13 +384,16 @@ 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, repoPath) => {
+    execute: async ({ input, repoPath, }) => {
         return strReplaceEditorExecutor({
-            command: "str_replace",
-            path: input.path,
-            old_str: input.old_str,
-            new_str: input.new_str,
-        }, repoPath);
+            input: {
+                command: "str_replace",
+                path: input.path,
+                old_str: input.old_str,
+                new_str: input.new_str,
+            },
+            repoPath,
+        });
     },
 };
 const stringInsertTool = {
@@ -403,13 +412,16 @@ 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."),
         }),
     },
-    execute: async (input, repoPath) => {
+    execute: async ({ input, repoPath, }) => {
         return strReplaceEditorExecutor({
-            command: "insert",
-            path: input.path,
-            insert_line: input.insert_line,
-            new_str: input.new_str,
-        }, repoPath);
+            input: {
+                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;AA2FnD,eAAO,MAAM,4BAA4B,EAAE,IA8G1C,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;AA4FnD,eAAO,MAAM,4BAA4B,EAAE,IA+H1C,CAAC"}

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

@@ -5,7 +5,8 @@ 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 fs_1 = __importDefault(require("fs"));
+const path_1 = __importDefault(require("path"));
 const zod_1 = require("zod");
 const run_1 = require("../agent/browsing/run");
 const utils_1 = require("../agent/browsing/utils");
@@ -88,12 +89,12 @@ exports.generateTestWithBrowserAgent = {
         description: BROWSER_AGENT_DESCRIPTION,
         parameters: BrowserAgentSchema,
     },
-    execute: async (input, repoPath, apiKey, trace, collectArtifacts) => {
+    execute: async ({ input, repoPath, trace, collectArtifacts, }) => {
         const repoDir = repoPath;
         const { testName, testSuites, fileName, changeToMake, project } = input;
+        const absoluteFilePath = path_1.default.join(repoDir, fileName);
         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,
@@ -108,20 +109,30 @@ exports.generateTestWithBrowserAgent = {
                 result: `Error reading playwright config: ${error}`,
             };
         }
-        if (!(0, web_1.hasTestBlock)({ testName, testSuites, filePath: fileName })) {
+        if (!fs_1.default.existsSync(absoluteFilePath)) {
+            return {
+                isError: true,
+                result: `Could not find file in repository: ${fileName}. Did you use the full path relative to the root of the repository?`,
+            };
+        }
+        if (!(0, web_1.hasTestBlock)({
+            testName,
+            testSuites,
+            filePath: absoluteFilePath,
+        })) {
             return {
                 isError: true,
                 result: `Test block not found for test name: "${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");
+        const fileBackup = fs_1.default.readFileSync(absoluteFilePath, "utf-8");
         try {
-            await (0, utils_1.replaceTodoWithCreateTest)(fileName);
+            await (0, utils_1.replaceTodoWithCreateTest)(fileName, repoDir);
         }
         catch (error) {
             // Undo the TODO -> createTest and test.only changes
-            await promises_1.default.writeFile(fileName, fileBackup, "utf-8");
+            fs_1.default.writeFileSync(absoluteFilePath, fileBackup, "utf-8");
             return {
                 isError: true,
                 result: `Error running tool: ${error}`,
@@ -150,7 +161,7 @@ exports.generateTestWithBrowserAgent = {
         });
         // Cleanup: Undo the TODO -> createTest changes
         await (0, pw_pause_1.revertToOriginalPwCode)(repoDir);
-        await promises_1.default.writeFile(fileName, fileBackup, "utf-8");
+        fs_1.default.writeFileSync(absoluteFilePath, fileBackup, "utf-8");
         const { isError, error, actionsSummary, artifacts } = toolResult;
         if (artifacts) {
             void collectArtifacts?.(artifacts);

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,IA6HpC,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,IA+HpC,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, repoPath, apiKey) => {
+    execute: async ({ input, apiKey, }) => {
         const { testRunUrl } = input;
         // Remove query parameters if they exist
         const urlWithoutParams = testRunUrl.split("?")[0] || testRunUrl;

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

package/dist/tools/test-run.js

@@ -34,7 +34,7 @@ exports.runTestTool = {
         description: "Run a test",
         parameters: RunTestSchema,
     },
-    execute: async (input, repoPath, apiKey, trace, collectArtifacts) => {
+    execute: async ({ input, repoPath, collectArtifacts, }) => {
         let reportUrl = undefined;
         let envOverrides = undefined;
         if ((0, artifacts_1.isArtifactCollectionEnabled)()) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.63.0",
+  "version": "0.64.0",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -62,7 +62,7 @@
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
     "zod": "^3.23.8",
-    "@empiricalrun/llm": "^0.17.1",
+    "@empiricalrun/llm": "^0.17.2",
     "@empiricalrun/r2-uploader": "^0.3.9",
     "@empiricalrun/test-run": "^0.9.4"
   },

package/tsconfig.tsbuildinfo

@@ -1 +1 @@
-{"root":["./src/index.ts","./src/actions/assert.ts","./src/actions/click.ts","./src/actions/done.ts","./src/actions/fill.ts","./src/actions/goto.ts","./src/actions/hover.ts","./src/actions/index.ts","./src/actions/next-task.ts","./src/actions/press.ts","./src/actions/skill.ts","./src/actions/text-content.ts","./src/actions/constants/index.ts","./src/actions/utils/index.ts","./src/agent/browsing/index.ts","./src/agent/browsing/run.ts","./src/agent/browsing/utils.ts","./src/agent/chat/agent-loop.ts","./src/agent/chat/exports.ts","./src/agent/chat/index.ts","./src/agent/chat/models.ts","./src/agent/chat/prompt.ts","./src/agent/chat/repo.ts","./src/agent/chat/state.ts","./src/agent/chat/types.ts","./src/agent/chat/utils.ts","./src/agent/codegen/create-test-block.ts","./src/agent/codegen/fix-ts-errors.ts","./src/agent/codegen/generate-code-apply-changes.ts","./src/agent/codegen/lexical-scoped-vars.ts","./src/agent/codegen/repo-edit.ts","./src/agent/codegen/run.ts","./src/agent/codegen/skills-retriever.ts","./src/agent/codegen/test-update-feedback.ts","./src/agent/codegen/types.ts","./src/agent/codegen/update-flow.ts","./src/agent/codegen/use-skill.ts","./src/agent/codegen/utils.ts","./src/agent/cua/computer.ts","./src/agent/cua/index.ts","./src/agent/cua/model.ts","./src/agent/cua/pw-codegen/element-from-point.ts","./src/agent/cua/pw-codegen/types.ts","./src/agent/cua/pw-codegen/pw-pause/index.ts","./src/agent/cua/pw-codegen/pw-pause/utils.ts","./src/agent/diagnosis-agent/index.ts","./src/agent/diagnosis-agent/strict-mode-violation.ts","./src/agent/enrich-prompt/index.ts","./src/agent/enrich-prompt/utils.ts","./src/agent/infer-agent/index.ts","./src/agent/master/action-tool-calls.ts","./src/agent/master/element-annotation.ts","./src/agent/master/execute-browser-action.ts","./src/agent/master/execute-skill-action.ts","./src/agent/master/next-action.ts","./src/agent/master/planner.ts","./src/agent/master/run.ts","./src/agent/master/scroller.ts","./src/agent/master/with-hints.ts","./src/agent/master/browser-tests/cua.spec.ts","./src/agent/master/browser-tests/fixtures.ts","./src/agent/master/browser-tests/index.spec.ts","./src/agent/master/browser-tests/skills.spec.ts","./src/agent/master/icon-descriptor/index.ts","./src/agent/master/icon-descriptor/normalize-svg.ts","./src/agent/planner/run-time-planner.ts","./src/agent/planner/run.ts","./src/artifacts/index.ts","./src/bin/index.ts","./src/bin/logger/index.ts","./src/bin/utils/context.ts","./src/bin/utils/index.ts","./src/bin/utils/fs/index.ts","./src/bin/utils/platform/web/index.ts","./src/bin/utils/platform/web/test-files/ts-path-import-validate.ts","./src/bin/utils/scenarios/index.ts","./src/browser-injected-scripts/annotate-elements.spec.ts","./src/constants/index.ts","./src/errors/index.ts","./src/evals/add-scenario-agent.evals.ts","./src/evals/append-create-test-agent.evals.ts","./src/evals/fetch-pom-skills-agent.evals.ts","./src/evals/infer-master-or-code-agent.evals.ts","./src/evals/master-agent.evals.ts","./src/evals/type.ts","./src/evals/update-scenario-agent.evals.ts","./src/file/client.ts","./src/file/server.ts","./src/human-in-the-loop/cli.ts","./src/human-in-the-loop/index.ts","./src/human-in-the-loop/ipc.ts","./src/page/index.ts","./src/prompts/lib/ts-transformer.ts","./src/reporter/index.ts","./src/reporter/lib.ts","./src/session/index.ts","./src/test-build/index.ts","./src/tool-call-service/index.ts","./src/tool-call-service/utils.ts","./src/tools/codegen-agent.ts","./src/tools/commit-and-create-pr.ts","./src/tools/diagnosis-fetcher.ts","./src/tools/download-build.ts","./src/tools/environment-crud.ts","./src/tools/str_replace_editor.ts","./src/tools/test-gen-browser.ts","./src/tools/test-run.ts","./src/tools/grep/index.ts","./src/tools/grep/ripgrep/index.ts","./src/tools/grep/ripgrep/types.ts","./src/tools/test-run-fetcher/index.ts","./src/tools/test-run-fetcher/types.ts","./src/tools/utils/index.ts","./src/types/handlebars.d.ts","./src/types/index.ts","./src/uploader/index.ts","./src/uploader/utils.ts","./src/utils/checkpoint.ts","./src/utils/env.ts","./src/utils/exec.ts","./src/utils/file-tree.ts","./src/utils/file.ts","./src/utils/git.ts","./src/utils/html.ts","./src/utils/index.ts","./src/utils/repo-tree.ts","./src/utils/slug.ts","./src/utils/string.ts","./src/utils/stripAnsi.ts"],"version":"5.8.3"}
+{"root":["./src/index.ts","./src/actions/assert.ts","./src/actions/click.ts","./src/actions/done.ts","./src/actions/fill.ts","./src/actions/goto.ts","./src/actions/hover.ts","./src/actions/index.ts","./src/actions/next-task.ts","./src/actions/press.ts","./src/actions/skill.ts","./src/actions/text-content.ts","./src/actions/constants/index.ts","./src/actions/utils/index.ts","./src/agent/browsing/index.ts","./src/agent/browsing/run.ts","./src/agent/browsing/utils.ts","./src/agent/chat/agent-loop.ts","./src/agent/chat/exports.ts","./src/agent/chat/index.ts","./src/agent/chat/models.ts","./src/agent/chat/prompt.ts","./src/agent/chat/repo.ts","./src/agent/chat/state.ts","./src/agent/chat/types.ts","./src/agent/chat/utils.ts","./src/agent/codegen/create-test-block.ts","./src/agent/codegen/fix-ts-errors.ts","./src/agent/codegen/generate-code-apply-changes.ts","./src/agent/codegen/lexical-scoped-vars.ts","./src/agent/codegen/repo-edit.ts","./src/agent/codegen/run.ts","./src/agent/codegen/skills-retriever.ts","./src/agent/codegen/test-update-feedback.ts","./src/agent/codegen/types.ts","./src/agent/codegen/update-flow.ts","./src/agent/codegen/use-skill.ts","./src/agent/codegen/utils.ts","./src/agent/cua/computer.ts","./src/agent/cua/index.ts","./src/agent/cua/model.ts","./src/agent/cua/pw-codegen/element-from-point.ts","./src/agent/cua/pw-codegen/types.ts","./src/agent/cua/pw-codegen/pw-pause/index.ts","./src/agent/cua/pw-codegen/pw-pause/utils.ts","./src/agent/diagnosis-agent/index.ts","./src/agent/diagnosis-agent/strict-mode-violation.ts","./src/agent/enrich-prompt/index.ts","./src/agent/enrich-prompt/utils.ts","./src/agent/infer-agent/index.ts","./src/agent/master/action-tool-calls.ts","./src/agent/master/element-annotation.ts","./src/agent/master/execute-browser-action.ts","./src/agent/master/execute-skill-action.ts","./src/agent/master/next-action.ts","./src/agent/master/planner.ts","./src/agent/master/run.ts","./src/agent/master/scroller.ts","./src/agent/master/with-hints.ts","./src/agent/master/browser-tests/cua.spec.ts","./src/agent/master/browser-tests/fixtures.ts","./src/agent/master/browser-tests/index.spec.ts","./src/agent/master/browser-tests/skills.spec.ts","./src/agent/master/icon-descriptor/index.ts","./src/agent/master/icon-descriptor/normalize-svg.ts","./src/agent/planner/run-time-planner.ts","./src/agent/planner/run.ts","./src/artifacts/index.ts","./src/bin/index.ts","./src/bin/logger/index.ts","./src/bin/utils/context.ts","./src/bin/utils/index.ts","./src/bin/utils/fs/index.ts","./src/bin/utils/platform/web/index.ts","./src/bin/utils/platform/web/test-files/ts-path-import-validate.ts","./src/bin/utils/scenarios/index.ts","./src/browser-injected-scripts/annotate-elements.spec.ts","./src/constants/index.ts","./src/errors/index.ts","./src/evals/add-scenario-agent.evals.ts","./src/evals/append-create-test-agent.evals.ts","./src/evals/fetch-pom-skills-agent.evals.ts","./src/evals/infer-master-or-code-agent.evals.ts","./src/evals/master-agent.evals.ts","./src/evals/type.ts","./src/evals/update-scenario-agent.evals.ts","./src/file/client.ts","./src/file/server.ts","./src/human-in-the-loop/cli.ts","./src/human-in-the-loop/index.ts","./src/human-in-the-loop/ipc.ts","./src/page/index.ts","./src/prompts/lib/ts-transformer.ts","./src/reporter/index.ts","./src/reporter/lib.ts","./src/session/index.ts","./src/test-build/index.ts","./src/tool-call-service/index.ts","./src/tool-call-service/utils.ts","./src/tools/commit-and-create-pr.ts","./src/tools/diagnosis-fetcher.ts","./src/tools/download-build.ts","./src/tools/environment-crud.ts","./src/tools/str_replace_editor.ts","./src/tools/test-gen-browser.ts","./src/tools/test-run.ts","./src/tools/grep/index.ts","./src/tools/grep/ripgrep/index.ts","./src/tools/grep/ripgrep/types.ts","./src/tools/test-run-fetcher/index.ts","./src/tools/test-run-fetcher/types.ts","./src/tools/utils/index.ts","./src/types/handlebars.d.ts","./src/types/index.ts","./src/uploader/index.ts","./src/uploader/utils.ts","./src/utils/checkpoint.ts","./src/utils/env.ts","./src/utils/exec.ts","./src/utils/file-tree.ts","./src/utils/file.ts","./src/utils/git.ts","./src/utils/html.ts","./src/utils/index.ts","./src/utils/repo-tree.ts","./src/utils/slug.ts","./src/utils/string.ts","./src/utils/stripAnsi.ts"],"version":"5.8.3"}

package/dist/tools/codegen-agent.d.ts

@@ -1,3 +0,0 @@
-import type { Tool } from "@empiricalrun/llm/chat";
-export declare const codegenTool: Tool;
-//# sourceMappingURL=codegen-agent.d.ts.map

package/dist/tools/codegen-agent.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"codegen-agent.d.ts","sourceRoot":"","sources":["../../src/tools/codegen-agent.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAmBnD,eAAO,MAAM,WAAW,EAAE,IAyBzB,CAAC"}

package/dist/tools/codegen-agent.js

@@ -1,40 +0,0 @@
-"use strict";
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.codegenTool = void 0;
-const zod_1 = require("zod");
-const run_1 = require("../agent/codegen/run");
-const CodegenSchema = zod_1.z.object({
-    testName: zod_1.z.string().describe("The name of the test to create or modify"),
-    testSuites: zod_1.z
-        .array(zod_1.z.string())
-        .describe("The suites (describe blocks) where the test is located"),
-    fileName: zod_1.z
-        .string()
-        .describe("The name of the file where the test is located. File name must end with .spec.ts"),
-    changeToMake: zod_1.z.string().describe("The change to make to the test"),
-});
-exports.codegenTool = {
-    schema: {
-        name: "generateTestWithCodegen",
-        description: "Create or modify a test case with code generation. This is useful when modifications can be done with TypeScript only, and don't require any browser interactions or element selectors.",
-        parameters: CodegenSchema,
-    },
-    execute: async (input) => {
-        const { testName, testSuites, fileName, changeToMake } = input;
-        const testCase = {
-            id: 0,
-            name: testName,
-            filePath: fileName,
-            suites: testSuites,
-            steps: [changeToMake],
-        };
-        const result = await (0, run_1.generateTestWithCodegen)({
-            testCase,
-            file: fileName,
-        });
-        return {
-            result: JSON.stringify(result),
-            isError: false,
-        };
-    },
-};