It's our intention for s&box to be cross-platform, a native Linux client is a good start.

We have Low/Med/High graphics settings but default to High, we should automatically select one based on the user's hardware.

Make it easier to make UIs that can be used with controller.

A small unnoticable compile step for scene map meshes that merges them together into one.

We want to remove the native material editor, simplify it all and put it in the inspector.

We want to replace Qt with Avalonia UI for the editor UI, a native C# solution that should be easier to work with and faster.

Support importing glTF models into the editor.

We don't have a good way of inspecting all resources currently resident on the GPU. We want to be able to quickly glance and filter down what our GPU memory is being used for.

Material changes are expensive to our frame time, we can avoid rebinding textures between materials, which in turn lets us instance more stuff together.

Skinned meshes currently can't be instanced, meaning every animated character is a separate draw call. We want to support GPU skinning with instancing so crowds and large numbers of animated entities are significantly cheaper to render. 

Replace our current shader compiler DXC with Slang, a modern shader language with features like interfaces, generics, and modules. This gives us a more expressive and maintainable shader codebase with better tooling support. 

Decrease the cost of rendering shadows by caching the static geometry and only rendering dynamic geometry over it.

We can break static meshes into meshlets / fragments and render them very efficiently, combining draw calls into indirect multi draw calls as well as using GPU occlusion culling. A static map or level can be entirely composed of these for huge benefit.

Terrain is currently rendered with one big geometry clipmap mesh. We want to split it up into a level of instanced clipmaps for a simple level of CPU culling, reducing GPU overhead especially in auxiliary views.

Instead of having to place volumes and configure them, we should be able to voxelize the entire scene (within reason) to figure out where to place probes.

Reintroduce multiview instancing for VR rendering, but using VK_KHR_multiview to make it transparent to the rendering pipeline.

A resource that goes hand-in-hand with Sprite, allowing you to define Tiles from a Tileset image so they can be painted into a scene. You will also be able to apply metadata like tags to tiles for things like custom footstep sounds on different materials.

This tool will let you paint Tiles into your scene by creating TilesetRenderers for each layer and giving you options for collision, depth, ect.

This will let you define the way that certain Tiles connect to one-another so you can paint with an Autotile brush and it will automatically place the correct tiles given the surrounding tiles.

Similar to Models, this will let you define attachment points on your sprite, adjusting it's position for each frame in an animation. This can be used to attach hats to a Head attachment, or having projectiles fire from a Muzzle attachment.

For animations that have a lead-in and hold on a pose/loop. Instead of splitting them up and using code to trigger them in time, you can use Loop Points to have a single animation with the lead-in and loop and when that animation is played the lead-in will only be played a single time.

A better performing animgraph system, where in our current system pose evaluation is expensive and fragile.

s&box's environmental audio simulation is pretty bare-bones. We only have very basic sound occlusion and static DSP volumes.

We want to add simulation for at least some of the physical sound properties:
reflection, transmition, absorbtion, diffraction...