Writing Good Test Cases

How to write test cases that the agent can execute reliably.

Structure

A test case consists of:

• Title — short human-readable name.
• Flows — one or more steps. Each flow has instructions (what to do) and an optional expected result (what to verify).

App context

App context is shared description injected into every case. Configure it once on the app:

• General rules — only non-obvious guidance (e.g. “Always grant permission prompts”).
• Credentials — test account email / password.
• Screen and feature names — define vocabulary once.
• Core functionality — what the app does.
• Game rules — win/lose conditions and key HUD elements for games.

App context

Login: [e-mail] / Test1234.

Home — first screen after login; search at top, recent items below.

Profile — bottom tab; name, avatar, settings.

Case instructions

Sign in and open Profile.

On Home, search for “shoes”.

Instructions

Don’t describe every action

The agent understands goals.

Avoid

Tap Next on “Welcome”. Tap Next on “Discover features”. Tap Get Started on “Ready?”.

Prefer

Complete onboarding until you reach the paywall.

Be specific when the agent goes off track

If a run wanders, name the element, screen, or break the flow into smaller steps.

Write step by step

One goal per numbered step — easier to debug and reuse as flows.

Expected results

State what should be true after the flow (visible screen, value, message). Vague “it works” expectations make failures harder to diagnose.

Reusable flows

Pull shared paths (login, onboarding) into reusable flows so many cases stay DRY. See Test Cases.