Best practices

Reliable tests use clear instructions, controlled data and small flows with one purpose.

Test one user journey at a time

Keep each test focused on one goal, such as signing up, creating an order or changing a password.

Short tests are easier to understand and rerun. Use folders to group related journeys instead of putting an entire app into one test.

Write clear steps

Tell the agent what to do, where to do it and what result proves the step passed.

• Use the visible names of pages, buttons and fields.

• Include exact text or values when they matter.

• Keep one action or check in each step.

• Describe the expected result instead of saying that it should “work”.

For example, use “Enter the email address, select Sign up, then check that the Dashboard appears” as separate steps rather than “Create an account”.

Reuse stable setup steps

Put repeated flows such as logging in or opening a project into reusable components.

Keep the component limited to setup that is shared by several tests. Product-specific actions should stay in the test so its purpose remains clear.

Use conditions for real alternatives

Add a condition when the app can validly show more than one path, not to hide an unreliable step.

Describe the condition using something the agent can see. Connect both outcomes when the test needs to handle either result.

Choose failure behaviour carefully

Use Stop on failure when later steps depend on the current step, such as login or navigation.

Use Continue on failure when later checks can still provide useful evidence. The failed step still affects the overall result.

Control users and data

Use dedicated test accounts with predictable permissions and data.

• Keep separate accounts for Live and development data.

• Give each account only the permissions needed by the test.

• Avoid real customer accounts and personal data.

• Reset or recreate records when a test depends on a known starting state.

Choose the right browser settings

Match the viewport to the device the journey is designed for.

Set a timeout long enough for the complete flow, including any waits or uploads. A very long timeout can make a stuck test slower to diagnose.

Use files only when needed

Attach only the files the agent needs for the test.

Use stable sample files without private information. Replace a file when its contents no longer match the expected result.

Keep the model consistent

Use the same model when comparing results over time.

A stronger model can help with complex or less predictable journeys. Changing the model can also change how the agent interprets the same instruction.

Investigate failed runs

Read the failed step and inspect its screenshot before changing the test.

Rerun once when the cause is unclear. Repeated failure in the same place usually means the app changed, the starting data is wrong, or the instruction needs more detail.

Maintain tests after app changes

Update tests when you rename visible controls, change navigation or replace a user journey.

Remove outdated tests and components. Old tests create noise and make real regressions harder to find.