@empiricalrun/test-gen 0.35.4 → 0.35.5

package/CHANGELOG.md

@@ -1,5 +1,12 @@
 # @empiricalrun/test-gen
 
+## 0.35.5
+
+### Patch Changes
+
+- 8e18e5b: feat: add scenario code agent evals
+- d6f9de2: fix: add tests for annotation script
+
 ## 0.35.4
 
 ### Patch Changes

package/browser-injected-scripts/annotate-elements.spec.ts

@@ -0,0 +1,258 @@
+// @ts-nocheck
+
+import { test } from "@playwright/test";
+import path from "path";
+
+test("should annotate all links on empirical landing page", async ({
+  page,
+}) => {
+  await page.goto(
+    "https://assets-test.empirical.run/selector-hints-testing/dom-1.html",
+  );
+
+  await page.addScriptTag({
+    path: path.resolve(__dirname, "./annotate-elements.js"),
+  });
+
+  const annotations = await page.evaluate(() => {
+    const { annotations } = window.annotateClickableElements();
+
+    return Object.entries(annotations).map(([hint, config]) => ({
+      hint,
+      innerText: config.node.innerText,
+      tagName: config.node.tagName,
+      href: config.node.href,
+    }));
+  });
+
+  test.expect(annotations).toEqual([
+    {
+      hint: "A",
+      innerText: "Empirical",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/",
+    },
+    {
+      hint: "B",
+      innerText: "Blog",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/blog",
+    },
+    {
+      hint: "C",
+      innerText: "Contact us",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/contact",
+    },
+    {
+      hint: "D",
+      href: "https://dash.empirical.run/",
+      innerText: "Login ↗\n(opens in a new tab)",
+      tagName: "A",
+    },
+    {
+      hint: "E",
+      innerText: "Get early access",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/contact",
+    },
+    {
+      hint: "F",
+      innerText: "Playwright\n(opens in a new tab)",
+      tagName: "A",
+      href: "https://github.com/microsoft/playwright",
+    },
+    {
+      hint: "G",
+      innerText: "Meet with us",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/contact",
+    },
+    {
+      hint: "H",
+      innerText: "Privacy Policy",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/privacy.html",
+    },
+  ]);
+});
+
+test("should annotate all important items on quizizz page", async ({
+  page,
+}) => {
+  await page.goto(
+    "https://assets-test.empirical.run/selector-hints-testing/dom-2/index.html",
+  );
+
+  await page.addScriptTag({
+    path: path.resolve(__dirname, "./annotate-elements.js"),
+  });
+
+  const annotations = await page.evaluate(() => {
+    const { annotations } = window.annotateClickableElements();
+
+    return Object.entries(annotations).map(([hint, config]) => ({
+      hint,
+      innerText: config.node.innerText.toLowerCase().trim(),
+      tagName: config.node.tagName,
+      testId: config.node.getAttribute("data-testid"),
+      href: config.node.href,
+    }));
+  });
+
+  test
+    .expect(annotations.find((item) => item.innerText.includes("enter code")))
+    .toBeTruthy();
+  test
+    .expect(annotations.find((item) => item.innerText.includes("get help")))
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("create") &&
+          item.testId === "create-content-button",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("explore") &&
+          item.href === "https://quizizz.com/admin",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("library") &&
+          item.href === "https://quizizz.com/admin/my-library/createdByMe",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("reports") &&
+          item.href === "https://quizizz.com/admin/reports",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("classes") &&
+          item.href === "https://quizizz.com/admin/classes",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("accommodations") &&
+          item.href ===
+            "https://quizizz.com/admin/differentiation/accommodations",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("quizizz ai") &&
+          item.href === "https://quizizz.com/admin/quizizz-ai",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("start your free trial") &&
+          item.href === "https://quizizz.com/super-pricing",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("upgrade") &&
+          item.href === "https://quizizz.com/super-pricing?backto=/admin",
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "INPUT" &&
+          item.testId === "emphasized-search-bar-input",
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" &&
+          item.innerText.includes("verify details") &&
+          item.testId === "verify-profile-cta",
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" && item.innerText.includes("for you"),
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" && item.innerText.includes("assessments"),
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" && item.innerText.includes("lessons"),
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" &&
+          item.innerText.includes("interactive videos"),
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" && item.innerText.includes("passages"),
+      ),
+    )
+    .toBeTruthy();
+});

package/dist/agent/codegen/run.d.ts

@@ -1,4 +1,12 @@
 import { TraceClient } from "@empiricalrun/llm";
 import { TestCase, TestGenConfigOptions } from "../../types";
+export declare function getAddScenarioCompletion({ testCase, testFiles, pageFiles, testFilePath, trace, options, }: {
+    testCase: TestCase;
+    testFiles: string;
+    pageFiles: string;
+    testFilePath: string;
+    trace?: TraceClient;
+    options?: TestGenConfigOptions;
+}): Promise<string>;
 export declare function generateTest(testCase: TestCase, file: string, options: TestGenConfigOptions, trace?: TraceClient): Promise<TestCase[]>;
 //# sourceMappingURL=run.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAkC,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAkBhF,OAAO,EAAE,QAAQ,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAI7D,wBAAsB,YAAY,CAChC,QAAQ,EAAE,QAAQ,EAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,oBAAoB,EAC7B,KAAK,CAAC,EAAE,WAAW,GAClB,OAAO,CAAC,QAAQ,EAAE,CAAC,CA0GrB"}
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAkC,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAkBhF,OAAO,EAAE,QAAQ,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAI7D,wBAAsB,wBAAwB,CAAC,EAC7C,QAAQ,EACR,SAAS,EACT,SAAS,EACT,YAAY,EACZ,KAAK,EACL,OAAO,GACR,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,OAAO,CAAC,EAAE,oBAAoB,CAAC;CAChC,mBA4BA;AAED,wBAAsB,YAAY,CAChC,QAAQ,EAAE,QAAQ,EAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,oBAAoB,EAC7B,KAAK,CAAC,EAAE,WAAW,GAClB,OAAO,CAAC,QAAQ,EAAE,CAAC,CAgFrB"}

package/dist/agent/codegen/run.js

@@ -3,7 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.generateTest = void 0;
+exports.generateTest = exports.getAddScenarioCompletion = void 0;
 const llm_1 = require("@empiricalrun/llm");
 const fs_extra_1 = __importDefault(require("fs-extra"));
 const logger_1 = require("../../bin/logger");
@@ -12,6 +12,35 @@ const web_1 = require("../../bin/utils/platform/web");
 const constants_1 = require("../../constants");
 const fix_ts_errors_1 = require("./fix-ts-errors");
 const update_flow_1 = require("./update-flow");
+async function getAddScenarioCompletion({ testCase, testFiles, pageFiles, testFilePath, trace, options, }) {
+    const promptSpan = trace?.span({
+        name: "add-scenario-prompt",
+    });
+    const instruction = await (0, llm_1.getPrompt)("add-scenario", {
+        testFiles: testFiles,
+        pageFiles: pageFiles,
+        scenarioName: testCase.name,
+        scenarioSteps: testCase.steps.join("\n"),
+        scenarioFile: testFilePath,
+    });
+    promptSpan?.end({ output: { instruction } });
+    const llm = new llm_1.LLM({
+        trace,
+        provider: options?.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER,
+        defaultModel: options?.model || constants_1.DEFAULT_MODEL,
+        providerApiKey: constants_1.MODEL_API_KEYS[options?.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER],
+    });
+    const firstShotMessage = await llm.createChatCompletion({
+        messages: instruction,
+        modelParameters: {
+            ...constants_1.DEFAULT_MODEL_PARAMETERS,
+            ...options?.modelParameters,
+        },
+    });
+    let response = firstShotMessage?.content || "";
+    return response;
+}
+exports.getAddScenarioCompletion = getAddScenarioCompletion;
 async function generateTest(testCase, file, options, trace) {
     const logger = new logger_1.CustomLogger();
     if (!fs_extra_1.default.existsSync(file)) {
@@ -35,43 +64,19 @@ async function generateTest(testCase, file, options, trace) {
         name: "create-test",
         input: {
             testCase,
-            file,
-            options,
+            testFiles: codePrompt,
+            pageFiles: pomPrompt,
+            testFilePath: file,
         },
     });
-    createTestSpan?.event({
-        name: "collate-files-as-text",
-        output: {
-            codePrompt,
-            pomPrompt,
-            testFileContent,
-        },
-    });
-    const promptSpan = createTestSpan?.span({
-        name: "add-scenario-prompt",
-    });
-    const instruction = await (0, llm_1.getPrompt)("add-scenario", {
+    const response = await getAddScenarioCompletion({
+        testCase,
         testFiles: codePrompt,
         pageFiles: pomPrompt,
-        scenarioName: testCase.name,
-        scenarioSteps: testCase.steps.join("\n"),
-        scenarioFile: file,
-    });
-    promptSpan?.end({ output: { instruction } });
-    const llm = new llm_1.LLM({
-        trace,
-        provider: options.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER,
-        defaultModel: options.model || constants_1.DEFAULT_MODEL,
-        providerApiKey: constants_1.MODEL_API_KEYS[options.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER],
-    });
-    const firstShotMessage = await llm.createChatCompletion({
-        messages: instruction,
-        modelParameters: {
-            ...constants_1.DEFAULT_MODEL_PARAMETERS,
-            ...options.modelParameters,
-        },
+        testFilePath: file,
+        trace: createTestSpan,
+        options,
     });
-    let response = firstShotMessage?.content || "";
     logger.success("Test generated successfully!");
     const readWriteFileSpan = trace?.span({ name: "write-to-file" });
     let contents = fs_extra_1.default.readFileSync(file, "utf-8");

package/dist/agent/codegen/update-flow.d.ts

@@ -3,6 +3,15 @@ import { TestCase, TestGenConfigOptions } from "../../types";
 type UpdatedTestCase = TestCase & {
     updatedFiles: string[];
 };
+export declare function getUpdateTestCodeCompletion({ testCase, testFileContent, testFiles, pageFiles, testFilePath, trace, options, }: {
+    testCase: TestCase;
+    testFiles: string;
+    pageFiles: string;
+    testFilePath: string;
+    testFileContent: string;
+    trace?: TraceClient;
+    options?: TestGenConfigOptions;
+}): Promise<string>;
 export declare function updateTest(testCase: TestCase, file: string, options: TestGenConfigOptions | undefined, logging?: boolean, validate?: boolean, trace?: TraceClient): Promise<UpdatedTestCase[]>;
 export declare function appendCreateTestBlock({ testCase, file, options, trace, validateTypes, }: {
     testCase: TestCase;

package/dist/agent/codegen/update-flow.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"update-flow.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/update-flow.ts"],"names":[],"mappings":"AAAA,OAAO,EAKL,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAsB3B,OAAO,EAAE,QAAQ,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAI7D,KAAK,eAAe,GAAG,QAAQ,GAAG;IAChC,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB,CAAC;AAqIF,wBAAsB,UAAU,CAC9B,QAAQ,EAAE,QAAQ,EAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,oBAAoB,GAAG,SAAS,EACzC,OAAO,GAAE,OAAc,EACvB,QAAQ,GAAE,OAAc,EACxB,KAAK,CAAC,EAAE,WAAW,GAClB,OAAO,CAAC,eAAe,EAAE,CAAC,CAsG5B;AAED,wBAAsB,qBAAqB,CAAC,EAC1C,QAAQ,EACR,IAAI,EACJ,OAAO,EACP,KAAK,EACL,aAAoB,GACrB,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,oBAAoB,CAAC;IAC/B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB,GAAG,OAAO,CAAC,eAAe,EAAE,CAAC,CA+E7B"}
+{"version":3,"file":"update-flow.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/update-flow.ts"],"names":[],"mappings":"AAAA,OAAO,EAKL,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAsB3B,OAAO,EAAE,QAAQ,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAI7D,KAAK,eAAe,GAAG,QAAQ,GAAG;IAChC,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB,CAAC;AAqIF,wBAAsB,2BAA2B,CAAC,EAChD,QAAQ,EACR,eAAe,EACf,SAAS,EACT,SAAS,EACT,YAAY,EACZ,KAAK,EACL,OAAO,GACR,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;IACxB,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,OAAO,CAAC,EAAE,oBAAoB,CAAC;CAChC,GAAG,OAAO,CAAC,MAAM,CAAC,CA6ClB;AAED,wBAAsB,UAAU,CAC9B,QAAQ,EAAE,QAAQ,EAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,oBAAoB,GAAG,SAAS,EACzC,OAAO,GAAE,OAAc,EACvB,QAAQ,GAAE,OAAc,EACxB,KAAK,CAAC,EAAE,WAAW,GAClB,OAAO,CAAC,eAAe,EAAE,CAAC,CA6D5B;AAED,wBAAsB,qBAAqB,CAAC,EAC1C,QAAQ,EACR,IAAI,EACJ,OAAO,EACP,KAAK,EACL,aAAoB,GACrB,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,oBAAoB,CAAC;IAC/B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB,GAAG,OAAO,CAAC,eAAe,EAAE,CAAC,CA+E7B"}

package/dist/agent/codegen/update-flow.js

@@ -3,7 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.appendCreateTestBlock = exports.updateTest = void 0;
+exports.appendCreateTestBlock = exports.updateTest = exports.getUpdateTestCodeCompletion = void 0;
 const llm_1 = require("@empiricalrun/llm");
 const crypto_1 = __importDefault(require("crypto"));
 const fs_extra_1 = __importDefault(require("fs-extra"));
@@ -103,41 +103,8 @@ async function applyFileChanges({ validateTypes = true, trace, testCase, fileCha
         logger.success(`${fileChange.filePath} file formatted successfully!`);
     }));
 }
-async function updateTest(testCase, file, options, logging = true, validate = true, trace) {
-    const logger = new logger_1.CustomLogger({ useReporter: logging });
-    const context = await (0, context_1.contextForGeneration)(file);
-    const { codePrompt, pomPrompt, testFileContent } = context;
-    const generatedTestCases = [];
-    logger.logEmptyLine();
-    const session = (0, session_1.getSessionDetails)();
-    trace =
-        trace ||
-            llm_1.langfuseInstance?.trace({
-                name: "update-test",
-                id: crypto_1.default.randomUUID(),
-                release: session.version,
-                tags: [
-                    options?.metadata.projectName || "",
-                    options?.metadata.environment || "",
-                ].filter((s) => !!s),
-            });
-    const updateTestSpan = trace?.span({
-        name: "update-test",
-        input: {
-            testCase,
-            file,
-            options,
-        },
-    });
-    updateTestSpan?.event({
-        name: "collate-files-as-text",
-        output: {
-            codePrompt,
-            pomPrompt,
-            testFileContent,
-        },
-    });
-    const promptSpan = updateTestSpan?.span({
+async function getUpdateTestCodeCompletion({ testCase, testFileContent, testFiles, pageFiles, testFilePath, trace, options, }) {
+    const promptSpan = trace?.span({
         name: "update-scenario-prompt",
     });
     const promptName = "update-scenario";
@@ -154,16 +121,16 @@ async function updateTest(testCase, file, options, logging = true, validate = tr
         suites: testCase?.suites || [],
     });
     const instruction = await (0, llm_1.getPrompt)(promptName, {
-        testFiles: codePrompt,
-        pageFiles: pomPrompt,
+        testFiles: testFiles,
+        pageFiles: pageFiles,
         scenarioName,
         scenarioSteps: testCase.steps.join("\n"),
-        scenarioFile: file,
+        scenarioFile: testFilePath,
         currentScenarioCodeBlock,
     });
     promptSpan?.end({ output: { instruction } });
     const llm = new llm_1.LLM({
-        trace: updateTestSpan,
+        trace,
         provider: options?.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER,
         defaultModel: options?.model || constants_1.DEFAULT_MODEL,
         providerApiKey: constants_1.MODEL_API_KEYS[options?.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER],
@@ -176,6 +143,43 @@ async function updateTest(testCase, file, options, logging = true, validate = tr
         },
     });
     let response = firstShotMessage?.content || "";
+    return response;
+}
+exports.getUpdateTestCodeCompletion = getUpdateTestCodeCompletion;
+async function updateTest(testCase, file, options, logging = true, validate = true, trace) {
+    const logger = new logger_1.CustomLogger({ useReporter: logging });
+    const context = await (0, context_1.contextForGeneration)(file);
+    const { codePrompt, pomPrompt, testFileContent } = context;
+    const generatedTestCases = [];
+    logger.logEmptyLine();
+    const session = (0, session_1.getSessionDetails)();
+    trace =
+        trace ||
+            llm_1.langfuseInstance?.trace({
+                name: "update-test",
+                id: crypto_1.default.randomUUID(),
+                release: session.version,
+                tags: [
+                    options?.metadata.projectName || "",
+                    options?.metadata.environment || "",
+                ].filter((s) => !!s),
+            });
+    const request = {
+        testCase,
+        testFileContent,
+        testFiles: codePrompt,
+        pageFiles: pomPrompt,
+        testFilePath: file,
+        options,
+    };
+    const updateTestSpan = trace?.span({
+        name: "update-test",
+        input: request,
+    });
+    const response = await getUpdateTestCodeCompletion({
+        ...request,
+        trace: updateTestSpan,
+    });
     logger.success("Test generated successfully!");
     const fileChanges = (0, utils_1.extractTestUpdates)(response);
     await applyFileChanges({

package/dist/agent/master/run.js

@@ -40,7 +40,7 @@ async function getNextAction({ task, executedActions, failedActions, pageUrl, tr
         failedActions: failedActions.map((a) => a).join("\n"),
         executedActions: executedActions.map((a) => a).join("\n"),
         pageUrl,
-    }, useHints ? 16 : 14);
+    }, 18);
     // assuming there is only one user message in the prompt. if there is a change in langfuse prompt format, this will need to be updated
     const userMessage = promptMessages.filter((m) => m.role === "user")[0];
     const systemMessage = promptMessages.filter((m) => m.role === "system")[0];

package/dist/browser-injected-scripts/annotate-elements.spec.ts

@@ -0,0 +1,258 @@
+// @ts-nocheck
+
+import { test } from "@playwright/test";
+import path from "path";
+
+test("should annotate all links on empirical landing page", async ({
+  page,
+}) => {
+  await page.goto(
+    "https://assets-test.empirical.run/selector-hints-testing/dom-1.html",
+  );
+
+  await page.addScriptTag({
+    path: path.resolve(__dirname, "./annotate-elements.js"),
+  });
+
+  const annotations = await page.evaluate(() => {
+    const { annotations } = window.annotateClickableElements();
+
+    return Object.entries(annotations).map(([hint, config]) => ({
+      hint,
+      innerText: config.node.innerText,
+      tagName: config.node.tagName,
+      href: config.node.href,
+    }));
+  });
+
+  test.expect(annotations).toEqual([
+    {
+      hint: "A",
+      innerText: "Empirical",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/",
+    },
+    {
+      hint: "B",
+      innerText: "Blog",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/blog",
+    },
+    {
+      hint: "C",
+      innerText: "Contact us",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/contact",
+    },
+    {
+      hint: "D",
+      href: "https://dash.empirical.run/",
+      innerText: "Login ↗\n(opens in a new tab)",
+      tagName: "A",
+    },
+    {
+      hint: "E",
+      innerText: "Get early access",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/contact",
+    },
+    {
+      hint: "F",
+      innerText: "Playwright\n(opens in a new tab)",
+      tagName: "A",
+      href: "https://github.com/microsoft/playwright",
+    },
+    {
+      hint: "G",
+      innerText: "Meet with us",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/contact",
+    },
+    {
+      hint: "H",
+      innerText: "Privacy Policy",
+      tagName: "A",
+      href: "https://assets-test.empirical.run/privacy.html",
+    },
+  ]);
+});
+
+test("should annotate all important items on quizizz page", async ({
+  page,
+}) => {
+  await page.goto(
+    "https://assets-test.empirical.run/selector-hints-testing/dom-2/index.html",
+  );
+
+  await page.addScriptTag({
+    path: path.resolve(__dirname, "./annotate-elements.js"),
+  });
+
+  const annotations = await page.evaluate(() => {
+    const { annotations } = window.annotateClickableElements();
+
+    return Object.entries(annotations).map(([hint, config]) => ({
+      hint,
+      innerText: config.node.innerText.toLowerCase().trim(),
+      tagName: config.node.tagName,
+      testId: config.node.getAttribute("data-testid"),
+      href: config.node.href,
+    }));
+  });
+
+  test
+    .expect(annotations.find((item) => item.innerText.includes("enter code")))
+    .toBeTruthy();
+  test
+    .expect(annotations.find((item) => item.innerText.includes("get help")))
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("create") &&
+          item.testId === "create-content-button",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("explore") &&
+          item.href === "https://quizizz.com/admin",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("library") &&
+          item.href === "https://quizizz.com/admin/my-library/createdByMe",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("reports") &&
+          item.href === "https://quizizz.com/admin/reports",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("classes") &&
+          item.href === "https://quizizz.com/admin/classes",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("accommodations") &&
+          item.href ===
+            "https://quizizz.com/admin/differentiation/accommodations",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("quizizz ai") &&
+          item.href === "https://quizizz.com/admin/quizizz-ai",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("start your free trial") &&
+          item.href === "https://quizizz.com/super-pricing",
+      ),
+    )
+    .toBeTruthy();
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.innerText.includes("upgrade") &&
+          item.href === "https://quizizz.com/super-pricing?backto=/admin",
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "INPUT" &&
+          item.testId === "emphasized-search-bar-input",
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" &&
+          item.innerText.includes("verify details") &&
+          item.testId === "verify-profile-cta",
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" && item.innerText.includes("for you"),
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" && item.innerText.includes("assessments"),
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" && item.innerText.includes("lessons"),
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" &&
+          item.innerText.includes("interactive videos"),
+      ),
+    )
+    .toBeTruthy();
+
+  test
+    .expect(
+      annotations.find(
+        (item) =>
+          item.tagName === "BUTTON" && item.innerText.includes("passages"),
+      ),
+    )
+    .toBeTruthy();
+});

package/dist/evals/add-scenario-agent.evals.d.ts

@@ -0,0 +1,4 @@
+import { EvaluateFn } from "./type";
+declare const addScenarioCodeAgentEvaluate: EvaluateFn;
+export default addScenarioCodeAgentEvaluate;
+//# sourceMappingURL=add-scenario-agent.evals.d.ts.map

package/dist/evals/add-scenario-agent.evals.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"add-scenario-agent.evals.d.ts","sourceRoot":"","sources":["../../src/evals/add-scenario-agent.evals.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AAEpC,QAAA,MAAM,4BAA4B,EAAE,UAkBnC,CAAC;AAEF,eAAe,4BAA4B,CAAC"}

package/dist/evals/add-scenario-agent.evals.js

@@ -0,0 +1,23 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+const run_1 = require("../agent/codegen/run");
+const addScenarioCodeAgentEvaluate = async ({ item, trace }) => {
+    const { testCase, testFiles, pageFiles, testFilePath } = item.input;
+    const response = await (0, run_1.getAddScenarioCompletion)({
+        testCase,
+        testFiles,
+        pageFiles,
+        testFilePath,
+        trace,
+    });
+    return {
+        scores: [
+            {
+                name: "equality",
+                value: item.expectedOutput === response ? 1 : 0,
+            },
+        ],
+        output: response,
+    };
+};
+exports.default = addScenarioCodeAgentEvaluate;

package/dist/evals/update-scenario-agent.evals.d.ts

@@ -0,0 +1,4 @@
+import { EvaluateFn } from "./type";
+declare const updateScenarioCodeAgentEvaluate: EvaluateFn;
+export default updateScenarioCodeAgentEvaluate;
+//# sourceMappingURL=update-scenario-agent.evals.d.ts.map

package/dist/evals/update-scenario-agent.evals.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"update-scenario-agent.evals.d.ts","sourceRoot":"","sources":["../../src/evals/update-scenario-agent.evals.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AAUpC,QAAA,MAAM,+BAA+B,EAAE,UAiDtC,CAAC;AAEF,eAAe,+BAA+B,CAAC"}

package/dist/evals/update-scenario-agent.evals.js

@@ -0,0 +1,49 @@
+"use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+const js_levenshtein_1 = __importDefault(require("js-levenshtein"));
+const update_flow_1 = require("../agent/codegen/update-flow");
+const utils_1 = require("../agent/codegen/utils");
+const updateScenarioCodeAgentEvaluate = async ({ item, trace }) => {
+    const { testCase, testFiles, pageFiles, testFilePath, testFileContent } = item.input;
+    const response = await (0, update_flow_1.getUpdateTestCodeCompletion)({
+        testCase,
+        testFiles,
+        pageFiles,
+        testFilePath,
+        testFileContent,
+        trace,
+    });
+    const fileChanges = (0, utils_1.extractTestUpdates)(response);
+    const expectedFileChanges = (0, utils_1.extractTestUpdates)(item.expectedOutput);
+    const fileChangeCount = fileChanges.length;
+    const expectedFileChangeCount = expectedFileChanges.length;
+    const correctFilePathChanges = expectedFileChanges.every((ef) => fileChanges.some((f) => f.filePath === ef.filePath));
+    const distanceScores = [];
+    expectedFileChanges.forEach((ef) => fileChanges.forEach((f) => {
+        if (f.filePath === ef.filePath && f.newCode && ef.newCode) {
+            const maxLength = ef.newCode.length > f.newCode.length
+                ? ef.newCode.length
+                : f.newCode.length;
+            distanceScores.push(1 - (0, js_levenshtein_1.default)(f.newCode || "", ef.newCode || "") / maxLength);
+        }
+    }));
+    let score = 0;
+    if (fileChangeCount === expectedFileChangeCount && correctFilePathChanges) {
+        score = distanceScores.length
+            ? distanceScores.reduce((agg, s) => agg * s)
+            : 0;
+    }
+    return {
+        scores: [
+            {
+                name: "score",
+                value: score,
+            },
+        ],
+        output: response,
+    };
+};
+exports.default = updateScenarioCodeAgentEvaluate;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.35.4",
+  "version": "0.35.5",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -52,8 +52,10 @@
     "@types/detect-port": "^1.3.5",
     "@types/express": "^4.17.21",
     "@types/fs-extra": "^11.0.4",
+    "@types/js-levenshtein": "^1.1.3",
     "@types/lodash.isequal": "^4.5.8",
-    "@types/md5": "^2.3.5"
+    "@types/md5": "^2.3.5",
+    "js-levenshtein": "^1.1.6"
   },
   "scripts": {
     "dev": "tsc --build --watch",
@@ -61,6 +63,7 @@
     "clean": "tsc --build --clean",
     "lint": "eslint .",
     "test": "vitest run",
+    "e2e-test": "npx playwright test",
     "test:watch": "vitest",
     "test:watch-files": "vitest $0 --watch"
   }

package/playwright.config.ts

@@ -0,0 +1,5 @@
+import { defineConfig } from "@playwright/test";
+
+export default defineConfig({
+  testDir: "./browser-injected-scripts",
+});