agentic-executor-e2e

August 25, 2026 · View on GitHub

End-to-end tests for the Voice Studio. Playwright drives a real browser against a real Next.js server.

See the repo README for the stack as a whole, and agentic-executor for the app under test.

Status

The suite is a scaffold, and it is deliberately thin. The studio's UX is still moving: tabs, clip review, and the copilot panel are all being reshaped. A broad end-to-end suite written against a layout that is about to change costs more to maintain than it catches, so coverage waits until the UX settles.

Correctness is covered in the meantime by the layers that are stable: pytest over every API surface, and Jest over the front end's data layer.

One consequence to know before you run it: the single spec in src/ asserts an h1 of "Voice models", and the studio now renders "Voice Studio". It fails until it is rewritten with the rest of the suite.

How it runs

flowchart LR
    NX["nx e2e agentic-executor-e2e"] --> PW[Playwright]
    PW -->|webServer| DEV["nx run …:dev"]
    PW --> BR["chromium · firefox · webkit"]
    BR -->|baseURL| DEV

Playwright starts the dev server itself through its webServer block, and reuses one that is already running. So a local nx dev in another terminal makes the suite start faster, and nothing breaks if there is none.

The suite runs against three browser projects: Chromium, Firefox, and WebKit. Mobile and branded-browser projects are present but commented out.

A trace is collected on the first retry of a failing test. Open it with npx playwright show-trace <path>.

Commands

Run from the repo root, in PowerShell.

nx e2e agentic-executor-e2e                  # every project
nx e2e agentic-executor-e2e -- --ui          # the Playwright UI runner
nx e2e agentic-executor-e2e -- --project=chromium
nx e2e agentic-executor-e2e -- --headed

The e2e target is inferred by @nx/playwright, so this project carries no project.json. Its package.json declares one implicit dependency on the web app, which is what makes nx affected run these tests when the app changes.

Playwright's browsers install separately:

npx playwright install

Configuration

Variable	Default	Purpose
`BASE_URL`	`http://localhost:3000`	The origin under test

The default matches next dev, which is what the webServer block starts. The compose stack publishes the same app on http://localhost:4001, so point BASE_URL there to test a running container instead of a dev server. Set it to the deployed origin in CI.

The config is a .mts file on purpose. Node forces ESM for that extension regardless of the workspace type, Playwright routes .mts through its ESM loader rather than the CJS compile path, and Nx's native TypeScript strip loads it directly. Playwright's config loader discovers it by extension.

Writing a test

apps/agentic-executor-e2e/
├── playwright.config.mts
└── src/
    └── *.spec.ts

Anything matching src/**/*.spec.ts is collected.

Two things worth holding to when the suite grows:

Test the workflow, not the markup. The valuable paths here cross processes: queue a video and watch its phase advance over SSE, assign a speaker to a voice, ask the copilot to do the same thing and see the same cache update. Those are what a unit test cannot reach.
Prefer role and text queries over CSS selectors. The h1 assertion that now fails is the argument: it broke on a heading rename that changed no behaviour at all.