fix(ci): address PR review comments on E2E test split

- Remove @requires-ai tag from 'empty query submission is prevented' test
  since it only validates UI state (send button disabled) without needing
  a database connection or AI embeddings
- Check all three Azure secrets (API_KEY, API_BASE, API_VERSION) in the
  secrets detection step, not just the API key
  fail, preserving test signal for both shards
- Update AGENTS.md to mention fork PRs alongside Dependabot as contexts
  where secrets are unavailable

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: gkorland

.github/workflows/playwright.yml

@@ -29,7 +29,7 @@ jobs:
     - name: Check for AI secrets
       id: check-secrets
       run: |
-        if [ -n "$AZURE_API_KEY" ]; then
+        if [ -n "$AZURE_API_KEY" ] && [ -n "$AZURE_API_BASE" ] && [ -n "$AZURE_API_VERSION" ]; then
           echo "has-ai-secrets=true" >> "$GITHUB_OUTPUT"
           echo "AI secrets are available — full E2E suite will run"
         else
@@ -38,6 +38,8 @@ jobs:
         fi
       env:
         AZURE_API_KEY: ${{ secrets.AZURE_API_KEY }}
+        AZURE_API_BASE: ${{ secrets.AZURE_API_BASE }}
+        AZURE_API_VERSION: ${{ secrets.AZURE_API_VERSION }}
 
     # Setup Python
     - name: Set up Python
@@ -191,8 +193,9 @@ jobs:
         CI: true
 
     # Run AI-dependent Playwright tests (only when secrets are available)
+    # Use !cancelled() so AI tests still run even if non-AI tests fail
     - name: Run Playwright tests (AI)
-      if: steps.check-secrets.outputs.has-ai-secrets == 'true'
+      if: "!cancelled() && steps.check-secrets.outputs.has-ai-secrets == 'true'"
       run: npx playwright test --grep @requires-ai --reporter=list
       env:
         CI: true

AGENTS.md

@@ -136,7 +136,7 @@ See `.env.example` for the full list.
 
 GitHub Actions workflows (`.github/workflows/`):
 - **tests.yml** — unit tests + lint on push/PR to main/staging
-- **playwright.yml** — dedicated Playwright E2E suite; non-AI tests always run, `@requires-ai` tests only when LLM secrets are available
+- **playwright.yml** — dedicated Playwright E2E suite; non-AI tests always run, `@requires-ai` tests only when LLM secrets are available (secrets are unavailable for Dependabot and fork PRs)
 - **pylint.yml** — Python linting
 - **spellcheck.yml** — docs spellcheck
 - **publish-docker.yml** — build & push Docker image to DockerHub

e2e/tests/chat.spec.ts

@@ -141,14 +141,11 @@ test.describe('Chat Feature Tests', () => {
     expect(resultsVisible).toBeTruthy();
   });
 
-  test('empty query submission is prevented', { tag: '@requires-ai' }, async () => {
+  test('empty query submission is prevented', async () => {
     const homePage = await browser.createNewPage(HomePage, getBaseUrl(), 'e2e/.auth/user.json');
     await browser.setPageToFullScreen();
 
-    // Ensure database is connected (will skip if already connected)
-    await homePage.ensureDatabaseConnected(apiCall);
-
-    // Verify send button is disabled with empty input
+    // Verify send button is disabled with empty input (pure UI validation, no DB needed)
     const isSendButtonDisabled = await homePage.isSendQueryButtonDisabled();
     expect(isSendButtonDisabled).toBeTruthy();
   });