E2E Testing: Stable Journeys and Useful Failure Evidence

What it does

ECC’s E2E source uses Playwright to explain test organization, page objects, configuration and continuous integration. It emphasizes semantic selectors and observable conditions over arbitrary sleeps, and treats traces, screenshots and video as evidence for diagnosing failures. It also discusses flaky-test quarantine and critical financial workflows. The point is to preserve confidence in a user journey, not maximize the number of browser scripts. Existing test infrastructure should be reused, and a page object is useful only when it removes real repetition or keeps interaction details in one coherent place.

When to use it

• Protect login and account switching after a navigation redesign.
• Reproduce a checkout validation defect without submitting a real payment.
• Diagnose a test that fails only when an API response arrives slowly in CI.

Step-by-step workflow

1. Select one business-critical journey and define an observable success condition, test data and prohibited side effects.
2. Reuse the project’s browser setup; isolate accounts and fixtures so parallel runs do not mutate the same records.
3. Locate controls through roles and labels, waiting for meaningful states rather than fixed delays.
4. Capture traces and screenshots on failure, then classify product defects, environment failures and nondeterministic tests separately.
5. Run the focused journey in the intended CI context and document any quarantine with an owner and removal condition.

Example request

Add a focused Playwright check for the shipping-country validation bug. Use our sandbox store and a synthetic customer. Select the US, enter an intentionally invalid address, then correct it and verify the validation message clears. Stop before order placement or payment submission. Prefer labels and roles over CSS classes, control network timing where needed and retain a trace on failure. Do not add a new framework or hide a failing assertion behind retries. Report the exact environment and what the test does not prove.

Worked example

A checkout test passes locally because it clicks the next button before an asynchronous address check finishes. Adding a longer sleep merely shifts the timing assumption. Instead, the test observes completion of validation and asserts the visible error state before correcting the address. A deliberately slow response exercises the race deterministically. The successful run demonstrates form behavior in the sandbox, while the report explicitly leaves payment authorization and fulfillment untested. If the test fails, its trace should show both the response and the interaction sequence without exposing a real customer.

What you get

A focused browser test, isolated fixture setup, failure artifacts and a report distinguishing application failure from infrastructure or incomplete coverage.

How to evaluate the result

• The journey waits for observable application state rather than elapsed time.
• The check cannot place a real order or charge a real customer.
• Failures retain enough redacted evidence to reproduce the cause.

What you need

A safe test environment and an existing browser-test runner or an explicitly chosen setup. External services need sandbox accounts or controlled test doubles.

Setup and activation

The reviewed ECC source provides an .agents skill layout. Consult the repository README for the chosen client, retain referenced supporting files and explicitly activate only the workflow needed. This catalogue read source files; it installed and executed nothing. Client compatibility depends on available tools and instruction loading, and was not runtime-tested here. Before installation, scan the exact source version, review permissions and use disposable non-sensitive examples.

Limits and considerations

Retries can reduce visible failures without fixing nondeterminism. Quarantine removes coverage until repaired and needs ownership. Browser traces may capture sensitive data, so retention and access must be deliberate. A sandbox checkout never proves a live payment or delivery.

Related workflows

• Use Frontend Patterns to repair races and focus behavior.
• Use Eval Harness for repeated agent-task quality beyond a single browser journey.

License and reuse

MIT in the selected SKILL.md and repository LICENSE. Preserve notices; external packages, services and example dependencies retain their own terms.

Further reading

• Reviewed skill at immutable commit
• Reviewed publisher README
• Reviewed MIT license