Reference dist/ in examples and evals (#418)

* fix types in evals

* fix examples

* build stagehand in ci

Author: kamath

.github/workflows/ci.yml

@@ -98,6 +98,9 @@ jobs:
       - name: Install Playwright browsers
         run: npm exec playwright install --with-deps
 
+      - name: Build Stagehand
+        run: npm run build
+
       - name: Run E2E Tests (Deterministic Playwright)
         run: npm run e2e
 
@@ -132,6 +135,9 @@ jobs:
       - name: Install Playwright browsers
         run: npm exec playwright install --with-deps
 
+      - name: Build Stagehand
+        run: npm run build
+
       - name: Run E2E Tests (browserbase)
         run: npm run e2e:bb
 
@@ -160,6 +166,9 @@ jobs:
       - name: Install dependencies
         run: npm install --no-frozen-lockfile
 
+      - name: Build Stagehand
+        run: npm run build
+
       - name: Install Playwright browsers
         run: npm exec playwright install --with-deps
 
@@ -211,6 +220,9 @@ jobs:
       - name: Install Playwright browsers
         run: npm exec playwright install --with-deps
 
+      - name: Build Stagehand
+        run: npm run build
+
       - name: Run Act Evals
         run: npm run evals category act
 
@@ -258,6 +270,9 @@ jobs:
       - name: Install dependencies
         run: npm install --no-frozen-lockfile
 
+      - name: Build Stagehand
+        run: npm run build
+
       - name: Install Playwright browsers
         run: npm exec playwright install --with-deps
 
@@ -323,6 +338,9 @@ jobs:
       - name: Install Playwright browsers
         run: npm exec playwright install --with-deps
 
+      - name: Build Stagehand
+        run: npm run build
+
       # 1. Run text_extract category with textExtract first
       - name: Run text_extract Evals (textExtract)
         run: npm run evals category text_extract -- --extract-method=textExtract
@@ -386,6 +404,9 @@ jobs:
       - name: Install Playwright browsers
         run: npm exec playwright install --with-deps
 
+      - name: Build Stagehand
+        run: npm run build
+
       - name: Run Observe Evals
         run: npm run evals category observe
 

evals/args.ts

@@ -1,5 +1,5 @@
 import process from "process";
-import { EvalCategorySchema } from "../types/evals";
+import { EvalCategorySchema } from "@/types/evals";
 
 // Extract command-line arguments passed to this script.
 const args = process.argv.slice(2);

evals/deterministic/stagehand.config.ts

@@ -1,28 +1,16 @@
-import type { ConstructorParams, LogLine } from "../../lib";
+import { default as DefaultStagehandConfig } from "@/stagehand.config";
+import type { ConstructorParams } from "@/dist";
 import dotenv from "dotenv";
 dotenv.config({ path: "../../.env" });
 
 const StagehandConfig: ConstructorParams = {
+  ...DefaultStagehandConfig,
   env: "LOCAL" /* Environment to run Stagehand in */,
-  apiKey: process.env.BROWSERBASE_API_KEY /* API key for authentication */,
-  projectId: process.env.BROWSERBASE_PROJECT_ID /* Project identifier */,
   verbose: 1 /* Logging verbosity level (0=quiet, 1=normal, 2=verbose) */,
-  debugDom: true /* Enable DOM debugging features */,
   headless: true /* Run browser in headless mode */,
-  logger: (message: LogLine) =>
-    console.log(
-      `[stagehand::${message.category}] ${message.message}`,
-    ) /* Custom logging function */,
-  domSettleTimeoutMs: 30_000 /* Timeout for DOM to settle in milliseconds */,
   browserbaseSessionCreateParams: {
     projectId: process.env.BROWSERBASE_PROJECT_ID,
   },
   enableCaching: false /* Enable caching functionality */,
-  browserbaseSessionID:
-    undefined /* Session ID for resuming browserbase sessions */,
-  modelName: "gpt-4o" /* Name of the model to use */,
-  modelClientOptions: {
-    apiKey: process.env.OPENAI_API_KEY,
-  } /* Configuration options for the model client */,
 };
 export default StagehandConfig;

evals/deterministic/tests/BrowserContext/addInitScript.test.ts

@@ -1,6 +1,6 @@
 import { test, expect } from "@playwright/test";
-import { Stagehand } from "../../../../lib";
-import StagehandConfig from "../../stagehand.config";
+import { Stagehand } from "@/dist";
+import StagehandConfig from "@/evals/deterministic/stagehand.config";
 
 test.describe("StagehandContext - addInitScript", () => {
   test("should inject a script on the context before pages load", async () => {

evals/deterministic/tests/BrowserContext/cookies.test.ts

@@ -1,6 +1,6 @@
 import { test, expect } from "@playwright/test";
-import { Stagehand } from "../../../../lib"; // Adjust the relative path as needed
-import StagehandConfig from "../../stagehand.config";
+import { Stagehand } from "@/dist";
+import StagehandConfig from "@/evals/deterministic/stagehand.config";
 
 test.describe("StagehandContext - Cookies", () => {
   let stagehand: Stagehand;

evals/deterministic/tests/BrowserContext/page.test.ts

@@ -1,6 +1,6 @@
 import { test, expect } from "@playwright/test";
-import { Stagehand } from "../../../../lib";
-import StagehandConfig from "../../stagehand.config";
+import { Stagehand } from "@/dist";
+import StagehandConfig from "@/evals/deterministic/stagehand.config";
 
 import http from "http";
 import express from "express";

evals/deterministic/tests/BrowserContext/routing.test.ts

@@ -1,6 +1,6 @@
 import { test, expect } from "@playwright/test";
-import { Stagehand } from "../../../../lib";
-import StagehandConfig from "../../stagehand.config";
+import { Stagehand } from "@/dist";
+import StagehandConfig from "@/evals/deterministic/stagehand.config";
 
 import http from "http";
 import express from "express";

evals/deterministic/tests/Errors/apiKeyError.test.ts

@@ -1,6 +1,6 @@
 import { test, expect } from "@playwright/test";
-import { Stagehand } from "../../../../lib";
-import StagehandConfig from "../../stagehand.config";
+import { Stagehand } from "@/dist";
+import StagehandConfig from "@/evals/deterministic/stagehand.config";
 import { z } from "zod";
 
 test.describe("API key/LLMClient error", () => {

evals/deterministic/tests/browserbase/contexts.test.ts

@@ -1,7 +1,7 @@
 import Browserbase from "@browserbasehq/sdk";
-import { Stagehand } from "../../../../lib";
 import { expect, test } from "@playwright/test";
-import StagehandConfig from "../../stagehand.config";
+import StagehandConfig from "@/evals/deterministic/stagehand.config";
+import { Stagehand } from "@/dist";
 
 // Configuration
 const CONTEXT_TEST_URL = "https://docs.browserbase.com";

evals/deterministic/tests/browserbase/downloads.test.ts

@@ -1,7 +1,7 @@
 import { test, expect } from "@playwright/test";
 import AdmZip from "adm-zip";
-import StagehandConfig from "../../stagehand.config";
-import { Stagehand } from "../../../../lib";
+import StagehandConfig from "@/evals/deterministic/stagehand.config";
+import { Stagehand } from "@/dist";
 import Browserbase from "@browserbasehq/sdk";
 
 const downloadRe = /sandstorm-(\d{13})+\.mp3/;