This guide explains how authored content is edited and run. The surface authority defines the tool/product boundary; the representative-scene matrix defines the current labs, galleries, and automated proof.

Edit And Preview

Open the scene in BulletSketch. Play in Viewport starts the current authored scene in-process using the editor's scene-play session. The editor continues to own the document, viewport, focus, diagnostics, and preview lifecycle.

Use this path for rapid editing, selection/inspection, playtest diagnostics, and authoring previews.

Run Standalone

Run Standalone launches one external pixelbullet-scene-runner for the current authored scene or temporary snapshot. BulletSketch supplies the scene path, current runtime asset base, control directory, and applicable launch options.

Development discovery uses Bazel runfiles. Extracted editor bundles resolve the runner only as the sibling executable in build_output/bin. READY and STOP coordination use atomically published files in the control directory.

Run A Scene Directly

The generic target is:

//tools/scene_runner:pixelbullet-scene-runner

Runner-specific inputs are --scene, --asset-base, and --control-directory. The runner installs ScenePlayLayer with the hybrid ShadowedMeshWithSprites profile. It owns no lab, gallery, or product rules.

The four authored labs use BulletSketch or this generic runner and have no dedicated wrapper. A future product executable runs product sessions and does not depend on the generic runner.

Assets

Editable scene references use @assets/... and @shared/.... BulletSketch can edit product-owned authored assets without launching a full product as long as the active workspace supplies the corresponding asset base and registration context. Product launch integration is a separate optional workflow, not part of generic authored-scene execution.

Automated Proof

//tests/runtime:scene_runner_smoke_tests proves the generic executable and control protocol. Editor locator, storage, controller, and lifecycle tests prove Run Standalone orchestration. GPU-tagged standalone runner tests exercise the four authored labs. Focused scene-play suites carry deterministic gameplay and presentation assertions.