Best practices

Writing stable tests

Give each step one clear job.
Include success criteria in the details.
Prefer user-visible goals over fragile selector or layout instructions.
Add exact button labels or field names only when the agent needs that precision.
Use conditions when the flow should branch based on what appears in the app.
Use components for repeated setup flows like logging in or opening the same page.
Use folders to organize related tests and run groups of runnable tests together.
Assign separate Live and Test users when the environments use different data.
Use a vision-capable model that is reliable enough for the suite.
Attach files to tests when the agent needs uploads or fixtures.

Use automatic login when the test should begin already authenticated.

Use password login when the test is verifying the visible login flow.

Do not pre-authenticate tests that are meant to verify login, signup, logout, SSO, password reset, or other auth behavior.

The instruction may be too broad. Add the page, controls, inputs, and expected result.

If the product still works, the test may be too brittle. Make the step more goal-oriented and keep exact UI wording only where it matters.

Add more detail or split the step into smaller steps.

Check the assigned Live user or Test user. Live branch runs use the Live user. Other branch runs use the Test user.

No matching test user may be assigned, or the assigned user may be disabled.

A folder needs at least one runnable test. Components can be organized in folders, but components are not run directly.

Check the step's On failure setting. Continue on failure records the problem and keeps going.

Was this helpful?