Pinning

Pressing P toggles the selected entity through Canvas_Scene_Toggle_Selected_Pin().

When pinning starts, Canvas_Entity_Capture_Pin() records:

• the entity's current screen-space origin;
• its current screen-space size.

After camera input, Canvas_Scene_Sync_Pinned() clamps that screen rectangle to the viewport, converts its origin back into world coordinates, and divides its screen size by the current zoom. The entity therefore still participates in the world renderer and picking system, but appears fixed in the camera view while the camera pans or zooms. Pinned entities are marked with the generated Lucide pin vector, drawn through the same Raylib icon path as the gallery.

Dragging or activating a pinned entity recaptures its screen-space state. Notifications are the exception: they use the dedicated animated stack in Canvas_Scene_Update_Notification_Stack() and cannot be toggled with P.

Visible context

Canvas_Scene_Build_Visible_Context() treats the camera viewport as the context boundary:

1. Convert the screen viewport to world bounds.
2. Intersect each entity's semantic bounds with that viewport.
3. Skip dismissed notifications.
4. Append identity, type, position, size, pin state, and type-specific state.
5. Report visible count, bytes written, and truncation.

Zooming out expands the world-space viewport, so more entities become context. Panning removes entities that no longer intersect it. The serializer preserves scene order and concatenates all visible entries into one provisional text snapshot.

Examples of semantic state include text-area content/editing, dropdown selection/open state, switch value, selected calendar date or table row, scroll offset, image source, browser URL, and Lucide gallery scroll/renderer metadata.

When adding an entity, never leave it on the generic value=%d fallback if an agent would need richer meaning. Update the serializer and the viewport/context tests together.