Objects in the world don’t decide what a click means themselves — they just report input events to MouseManager, which delegates to whichever MouseState is currently active. This keeps “what does clicking the ground do right now” in one place instead of scattered across every clickable node.

MouseManager link

src/Manager/mouse_manager.gd is an autoload holding current_state: MouseState. Clickable things report discrete events — object_clicked, on_object_hover, on_ground_input_event — and any code that needs “what’s under the mouse right now” can call get_floor_hit_at_mouse() (a raycast filtered to the "floor" group). It also owns entity selection (select_entity, entity_selected signal) and swaps the OS cursor for the active targeting mode.

MouseState subclasses link

Defined under src/Manager/MouseState/, an abstract base (mouse_state.gd) defines enter/exit/process_state/handle_object_clicked/ handle_ground_input_event/handle_right_click/handle_cancel. Three concrete states exist:

• MouseStateDefault — the everyday state: click the ground to move, click a BestiaVisual to attack it, click an ItemVisual to loot it.
• MouseStateSkillTargeting — entered when the player picks a skill that needs a target. Branches on the skill’s target_type: GROUND/AOE_GROUND shows a floor-following indicator sized to the skill’s aoe_radius; ENEMY/FRIENDLY snaps to the closest matching Entity within SettingsManager.skill_target_snap_distance (holding Shift inverts the friend/enemy filter). Confirming calls ConnectionManager.activate_skill(skill_id, level, position, [entity_id]).
• MouseStateItemTargeting — the equivalent flow for a scripted item (ItemUse. on_targeting_click), entered from ItemResource.use_item when the item needs a target instead of firing immediately.

Worked example: click to attack link

sequenceDiagram
  participant Area as BestiaVisual Area3D
  participant MM as MouseManager
  participant State as MouseStateDefault
  participant CM as ConnectionManager
  participant Net as BnetSocket

  Area->>MM: object_clicked(self, event, position)
  MM->>State: handle_object_clicked(mgr, object, event, position)
  State->>State: object is BestiaVisual → select_entity(object)
  State->>CM: send_attack_entity(entity_id, 0, 1)
  CM->>Net: SendMessage(AttackEntityCMSG)

1. Input: the target’s Area3D picks up the click via Godot physics picking (bestia_visual.gd):

     func _on_area_3d_input_event(_camera, event, event_position, _normal, _shape_idx) -> void:
       MouseManager.object_clicked(self, event, event_position)
     

2. Dispatch: MouseManager.object_clicked() forwards to the active state’s handle_object_clicked.
3. Default-state handling (mouse_state_default.gd): confirms the click was the normal_action input action, that the target isn’t a plain Interactable, and if it’s a BestiaVisual, selects it and fires the attack:

     if object is BestiaVisual:
       mgr.select_entity(object)
       ConnectionManager.send_attack_entity(object.get_bestia_entity_id(), 0, 1)
     

4. Send: ConnectionManager.send_attack_entity() builds an AttackEntityCMSG and sends it — see Networking for how that reaches zone-server.

Skill-based attacks follow the same shape through MouseStateSkillTargeting instead of the default state, ending in ConnectionManager.activate_skill() /ActivateSkillCMSG rather than send_attack_entity().

MouseMarker link

src/Game/MouseMarker/mouse_marker.gd is the always-visible ground-tile cursor, snapped to the 1m tile under the mouse. Only shown in the default mouse state — targeting states replace it with their own indicator (AOECastIndicator/EntitySnapIndicator, see World & Terrain for how the underlying floor raycast works and UI Overview for VFX like the AOE indicator).

Camera link

A third-person orbit camera under src/Game/SpringArmCamera/, split into two scripts:

• camera_spring_arm.gd rotates a Node3D (yaw/pitch, clamped) while the right mouse button is held and dragged past a small pixel threshold. A clean press-and-release without dragging past that threshold instead fires MouseManager.right_clicked (context menu) and cancels any active skill/item targeting rather than rotating the camera. Scroll wheel adjusts SpringArm3D. spring_length for zoom.
• camera_follow.gd, on the actual Camera3D, lerps its position toward the spring arm’s position every frame — decoupling render-smoothing from the arm’s own collision-driven length changes (the spring arm itself can snap shorter instantly when it hits geometry; the camera node smooths that out visually).