UI tests

Galata-based Playwright tests that drive a real JupyterLab instance with the notebook-intelligence labextension installed.

Run intent: manual, developer-triggered

This suite is intentionally not wired into the GitHub Actions build.yml workflow. Each run boots a full JupyterLab process, real browser contexts, and is comparatively expensive. We keep it as a high-signal pre-merge check that the maintainer reaches for when:

• a PR touches a flow with a regression history (see the per-spec comments),
• you're refactoring the chat sidebar, the notebook generation toolbar, the cell-output toolbar, or the launcher,
• you want to lock down a bug fix before merging.

The jest unit suite (jlpm jest at the repo root) stays the default on-every-commit check. Treat this Galata suite as top-of-the-pyramid coverage.

What's covered

Spec	Flows
extension.spec.ts	Smoke: extension activates, chat sidebar opens.
chat-sidebar.spec.ts	Footer icons + labels, gear-icon title, prefix popover via typed /, slash-button toggle, workspace file picker open + close, regression #262 (Escape from picker search).
notebook-toolbar.spec.ts	Toolbar button renders, popover structure, regression #231 (textarea focus on open), submit gating, Escape + outside-click dismissal.
cell-output.spec.ts	Hover toolbar renders Explain + Ask on clean cells, Troubleshoot only on errored cells, click activates the hovered cell.
launcher.spec.ts	Regression #260/#268 (Coding Agent tiles must add / dispose dynamically on capability change, not rely on isVisible).

When a flow needs a real LLM round trip to pass we skip it rather than mock the network — these tests are about UI state, not model output. Tests that require a configured chat provider skip cleanly on a fresh Galata workspace.

Running locally

From the repository root:

# Install the extension into the dev environment first.
pip install -e .[test]
jlpm
jlpm build

# Install ui-tests dependencies and Playwright browsers.
cd ui-tests
jlpm
jlpm playwright install chromium

# Run the suite.
jlpm test

jlpm test:debug opens the Playwright inspector for stepwise debugging; jlpm test:update regenerates snapshots when an intentional UI change makes the existing reference image stale.

Layout

• playwright.config.ts — boots jupyter lab via webServer, points Playwright at http://localhost:8888/lab, enables traces + videos on failure, and retries twice on CI.
• jupyter_server_test_config.py — disables auth/XSRF and pins the port so Playwright connects deterministically.
• tests/ — *.spec.ts files. Galata's test/expect come from @jupyterlab/galata and provide a page fixture that's already inside the lab shell.

Adding tests

Each spec file is independent. Use Galata's helpers (notebook commands, file browser, settings) before reaching for raw page.locator so tests stay resilient to lab-side DOM changes.