Task Tests
Use a task to play a game session, automate a workflow, document an app, or simulate a particular kind of user. Define what the agent needs to accomplish and the output you want.
Create a Task test in the Repository, select an agent, and set a minimum and maximum time budget for the whole task.
Task Instructions and Personality
| Field | What to write |
|---|---|
| Task instructions | The workflow, completion criteria, and requested deliverables. |
| Personality instructions | Optional guidance for how the agent behaves, such as a first-time user or a methodical UX reviewer. |
| Hints | Navigation or recovery guidance, available under show hints. |
For an app review:
Task instructions: Visit each screen in the onboarding flow. Document the navigation, required inputs, and confirmation messages. Export the findings as onboarding-review.pdf.
Personality instructions: Act as a first-time user. Read the on-screen guidance before choosing an action, and record points where the next action is unclear.
Attach knowledge files for report templates, questionnaires, or reference material. Give the output a filename and describe its contents in the task instructions.
For a game session:
Task instructions: Start a new game, complete the tutorial, and explore the first area. Try the available weapons and upgrades, then summarize your progress and any confusing moments.
Personality instructions: Play as a casual first-time player. Follow tutorial prompts and choose upgrades based on their in-game descriptions.
Review the Output
Open Run Details to review the execution and download generated files from artifacts. See Custom Output for supported deliverables and examples.
For exploratory bug finding, use Discovery. For repeatable pass/fail checks, use Verification.