Rig Template

• package.json scripts for build, start, and end-to-end tests
• playwright.config.* for baseURL, webServer, projects, and authentication
• next.config.* for existing experimental options
• CI, preview deployment, container, and hosting configuration
• test helpers for login, storageState, fixtures, flags, and seeded data

Ask only for details the repository cannot answer, such as unavailable credentials or which remote environment may expose the testing API.

What the rig must define

Production build and server

Use next build followed by next start, or a remote artifact produced by the same production build. Automatic prefetching does not run in next dev, so a development server cannot verify preservation.

Record separate build and start commands. For a local rig, record the port, stop any previous server before starting, fail on EADDRINUSE, and confirm the new process owns the port before running Playwright. next start can fork a next-server child, so the launcher process ID may not own the port. Start the server in a process group that the rig can stop as a unit, or discover and stop the process listening on the recorded port before the next build.

Testing API

An instant() test against a production build requires experimental.exposeTestingApiInProductionBuild. Gate it so real production builds do not expose the API:

Merge the option into an existing experimental object instead of replacing the project’s other experimental options.

Set the condition while running next build. Setting it only for next start is too late because the testing API is compiled into the production artifact. When the artifact was built without it, Next.js does not activate the navigation lock, so the test cannot distinguish prefetched UI from streamed dynamic content. Rebuild with the condition enabled before interpreting the results. Use the project’s existing environment naming when it already distinguishes test, staging, preview, and production builds.

Test command and base URL

Record the exact Playwright command and how it receives the measured build’s URL. Reuse the project’s package manager, Playwright configuration, projects, and reporters. The suite must import instant() from @next/playwright. If the dependencies are absent, install @next/playwright on the same release line as the project’s next, alongside @playwright/test.

For a local rig, a typical sequence is:

Adapt the script names and port to the project. Keep the production server running while the test command executes. Follow the public client-navigation test (opens in a new tab): load the source route, confirm the real Link is visible, then enter instant(), click, wait for the destination URL, and assert the prefetched UI.

Test context

Record the state required to reach the audited Links and destination UI:

• Use public; no authentication when the navigation is public.
• Otherwise record the test account and login mechanism, including a fixture, storageState, API login, or seeded session.
• Record flags, plan, role, locale, seeded data, and other state that can change which UI the test sees.

A test user is not required. The field exists to make authenticated and state-dependent tests reproducible when the app needs one.

Drift

List differences between the state used to choose the preservation target and the state used by Playwright. Feature flags, permissions, empty test data, and locale differences can make an assertion fail because the target is unreachable, not because Partial Prefetching removed it. Write none known only after checking the test context.

Iteration loop

Record the complete loop the agent can repeat without rediscovering commands:

• Local: build with the testing API, start the new artifact, run the focused suite, stop the server, edit, and repeat.
• Remote: push, wait for the measured artifact, verify it matches HEAD, run the focused suite against its URL, edit, and repeat.

Note any step the agent cannot perform without the user, including deployment approval, protected branches, secrets, or multi-factor authentication.

Artifact liveness

For a remote rig, record how the test proves the deployment matches HEAD. Prefer an endpoint or response header that exposes the deployed commit SHA. If the app has neither, use the deployment provider’s API to select the artifact whose commit SHA matches HEAD.

A freshly completed local build followed by start does not need a SHA probe. Record n/a; local build and start.

Walls

Record build and run obstacles with their working resolution, such as required environment variables, server-only imports that fail during prerendering, unavailable credentials, or a process that keeps reclaiming the test port. Reuse these notes on the next iteration.

Write instant-nav.rig.md

Place this file at the repository root or next to the end-to-end configuration:

Every field needs a concrete value. n/a is valid only with a reason, such as TEST USER: public; no authentication or LIVENESS: n/a; local build and start.

Check the rig before writing the baseline

Before recording the legacy prefetched UI:

1. Build with the testing API condition enabled.
2. Start or locate that exact artifact and confirm the base URL responds.
3. Run one focused instant() smoke test through a real <Link> navigation.
4. Confirm the test can reach its source Link and eventual destination UI in the recorded test context.

Fix the rig before interpreting a preservation failure. A missing testing API, stale deployment, unreachable target, or wrong test state is an environment failure rather than evidence that the migration changed the prefetch.