Run a PIE test

Before you start

• Hayba is installed and your agent is connected. See Getting started.
• Your level has something to test. The prompt below assumes a start button and a door Blueprint with a boolean bIsOpen; change the names to yours.
• This guide reads and drives Play, so it works with Review each external edit on. One exception: editor_pie_click_actor is held for approval, which the exact review cannot give on this version. Clicking widgets and pressing keys is not held.
• Compile your Blueprints first. Hayba will not start Play while a loaded Blueprint would open a compile-error dialog.

Prompt

Run a PIE test. Check the editor is idle, start Play-In-Editor, read what is on screen and click the button labelled "Start". Then check the actor BP_Door until its property bIsOpen is true, trying up to 10 times. Take a screenshot, stop Play, and tell me whether the test passed and what you observed.

What the agent does

1. editor_get_state confirms nothing is playing or building.
2. editor_start_pie starts Play. If a Blueprint needs fixing first, it answers pie_blocked and names it.
3. editor_pie_widget_tree lists the visible widgets with their text and screen rectangles.
4. editor_pie_click_widget clicks the control whose text matches "Start", and reports how many matched.
5. editor_pie_wait_for reads bIsOpen once per call. The agent repeats it until it matches or gives up.
6. editor_pie_screenshot requests a screenshot. The file appears a frame or two later, so the agent checks again with check_only.
7. editor_stop_pie ends the session.

Things worth knowing:

• A command cannot wait inside the game: it runs on the game thread, and the world does not advance while it runs. So editor_pie_wait_for and editor_pie_assert each take one observation. A result of polling: true means "not yet", and the agent asks again.
• editor_pie_press_key sends keyboard and gamepad buttons, editor_pie_type_text types into a focused field, and editor_pie_actor_list finds runtime actors with their exact paths.
• While Play runs, edits are refused with pie_active. Stop Play before asking for changes.
• Only the agent that started a session can drive it. Agents never drive a session you started.

Automation tests

To run Unreal automation tests instead of a scripted Play session:

1. test_list with filter_pattern finds tests by name.
2. test_run starts them as a background job and returns a job_id. Most tests run in a separate editor process, so the open editor stays responsive.
3. build_status reports progress and the final pass, fail and skip counts.
4. test_get_log returns the full errors and warnings of the last run.

A filter that matches no tests fails, rather than reporting an empty pass.

Check the result

• Ask the agent for each observation it made, not only a verdict.
• Screenshots land in Saved/Screenshots/<Platform> unless the agent gave an absolute path.
• editor_get_state should show Play as none when the test ends.