update tests

Author: filip-michalsky

.github/workflows/test.yml

@@ -106,12 +106,12 @@ jobs:
     
     - name: Run local integration tests
       run: |
-        # Run integration tests marked as 'local' and not 'slow'
+        # Run integration tests marked as 'integration' and 'local'
         xvfb-run -a pytest tests/integration/ -v \
           --cov=stagehand \
           --cov-report=xml \
           --junit-xml=junit-integration-local.xml \
-          -m "local and not slow" \
+          -m "integration and local" \
           --tb=short \
           --maxfail=5
       env:
@@ -190,14 +190,14 @@ jobs:
         name: integration-test-results-slow
         path: junit-integration-slow.xml
 
-  test-browserbase:
-    name: Browserbase Integration Tests
+  test-integration-api:
+    name: API Integration Tests
     runs-on: ubuntu-latest
     needs: test-unit
     if: |
       github.event_name == 'schedule' || 
-      contains(github.event.pull_request.labels.*.name, 'test-browserbase') ||
-      contains(github.event.pull_request.labels.*.name, 'browserbase')
+      contains(github.event.pull_request.labels.*.name, 'test-api') ||
+      contains(github.event.pull_request.labels.*.name, 'api')
     
     steps:
     - uses: actions/checkout@v4
@@ -215,26 +215,26 @@ jobs:
         # Install temporary Google GenAI wheel
         pip install temp/google_genai-1.14.0-py3-none-any.whl
     
-    - name: Run Browserbase tests
+    - name: Run API integration tests
       run: |
-        pytest tests/ -v \
+        pytest tests/integration/ -v \
           --cov=stagehand \
           --cov-report=xml \
-          --junit-xml=junit-browserbase.xml \
-          -m "browserbase" \
+          --junit-xml=junit-integration-api.xml \
+          -m "integration and api" \
           --tb=short
       env:
         BROWSERBASE_API_KEY: ${{ secrets.BROWSERBASE_API_KEY }}
         BROWSERBASE_PROJECT_ID: ${{ secrets.BROWSERBASE_PROJECT_ID }}
         MODEL_API_KEY: ${{ secrets.MODEL_API_KEY }}
         STAGEHAND_API_URL: ${{ secrets.STAGEHAND_API_URL }}
 
-    - name: Upload Browserbase test results
+    - name: Upload API integration test results
       uses: actions/upload-artifact@v4
       if: always()
       with:
-        name: browserbase-test-results
-        path: junit-browserbase.xml
+        name: api-integration-test-results
+        path: junit-integration-api.xml
 
   test-performance:
     name: Performance Tests
@@ -373,6 +373,11 @@ jobs:
       with:
         python-version: "3.11"
 
+    - name: Install system dependencies
+      run: |
+        sudo apt-get update
+        sudo apt-get install -y xvfb
+    
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
@@ -381,10 +386,11 @@ jobs:
         # Install temporary Google GenAI wheel
         pip install temp/google_genai-1.14.0-py3-none-any.whl
         playwright install chromium
+        playwright install-deps chromium
     
     - name: Run E2E tests
       run: |
-        pytest tests/ -v \
+        xvfb-run -a pytest tests/ -v \
           --cov=stagehand \
           --cov-report=xml \
           --junit-xml=junit-e2e.xml \
@@ -395,6 +401,7 @@ jobs:
         BROWSERBASE_PROJECT_ID: ${{ secrets.BROWSERBASE_PROJECT_ID || 'mock-project-id' }}
         MODEL_API_KEY: ${{ secrets.MODEL_API_KEY || 'mock-model-key' }}
         STAGEHAND_API_URL: ${{ secrets.STAGEHAND_API_URL || 'http://localhost:3000' }}
+        DISPLAY: ":99"
 
     - name: Upload E2E test results
       uses: actions/upload-artifact@v4

pytest.ini

@@ -9,6 +9,9 @@ markers =
     unit: marks tests as unit tests
     integration: marks tests as integration tests
     smoke: marks tests as smoke tests
+    local: marks tests as local integration tests
+    api: marks tests as API integration tests
+    e2e: marks tests as end-to-end tests
 
 log_cli = true
 log_cli_level = INFO 

tests/integration/api/test_core_api.py

@@ -30,6 +30,8 @@ async def stagehand_api():
 
 
 @skip_if_no_creds
+@pytest.mark.integration
+@pytest.mark.api
 @pytest.mark.asyncio
 async def test_stagehand_api_initialization(stagehand_api):
     """Ensure that Stagehand initializes correctly against the Browserbase API."""

tests/integration/local/test_core_local.py

@@ -14,6 +14,8 @@ async def stagehand_local():
     await sh.close()
 
 
+@pytest.mark.integration
+@pytest.mark.local
 @pytest.mark.asyncio
 async def test_stagehand_local_initialization(stagehand_local):
     """Ensure that Stagehand initializes correctly in LOCAL mode."""