diff --git a/.claude/CLAUDE.md b/.claude/CLAUDE.md
new file mode 120000
index 00000000..47dc3e3d
--- /dev/null
+++ b/.claude/CLAUDE.md
@@ -0,0 +1 @@
+AGENTS.md
\ No newline at end of file
diff --git a/.claude/rules/creating-rules.md b/.claude/rules/creating-rules.md
new file mode 120000
index 00000000..6fb51cc9
--- /dev/null
+++ b/.claude/rules/creating-rules.md
@@ -0,0 +1 @@
+../../.cursor/rules/creating-rules.mdc
\ No newline at end of file
diff --git a/.claude/rules/events.md b/.claude/rules/events.md
new file mode 120000
index 00000000..72ba2224
--- /dev/null
+++ b/.claude/rules/events.md
@@ -0,0 +1 @@
+../../.cursor/rules/events.mdc
\ No newline at end of file
diff --git a/.claude/rules/node-schemas.md b/.claude/rules/node-schemas.md
new file mode 120000
index 00000000..93ceffa0
--- /dev/null
+++ b/.claude/rules/node-schemas.md
@@ -0,0 +1 @@
+../../.cursor/rules/node-schemas.mdc
\ No newline at end of file
diff --git a/.claude/rules/renderers.md b/.claude/rules/renderers.md
new file mode 120000
index 00000000..a7bb563a
--- /dev/null
+++ b/.claude/rules/renderers.md
@@ -0,0 +1 @@
+../../.cursor/rules/renderers.mdc
\ No newline at end of file
diff --git a/.claude/rules/scene-registry.md b/.claude/rules/scene-registry.md
new file mode 120000
index 00000000..017d4c15
--- /dev/null
+++ b/.claude/rules/scene-registry.md
@@ -0,0 +1 @@
+../../.cursor/rules/scene-registry.mdc
\ No newline at end of file
diff --git a/.claude/rules/selection-managers.md b/.claude/rules/selection-managers.md
new file mode 120000
index 00000000..3c2d1d93
--- /dev/null
+++ b/.claude/rules/selection-managers.md
@@ -0,0 +1 @@
+../../.cursor/rules/selection-managers.mdc
\ No newline at end of file
diff --git a/.claude/rules/spatial-queries.md b/.claude/rules/spatial-queries.md
new file mode 120000
index 00000000..c3345522
--- /dev/null
+++ b/.claude/rules/spatial-queries.md
@@ -0,0 +1 @@
+../../.cursor/rules/spatial-queries.mdc
\ No newline at end of file
diff --git a/.claude/rules/systems.md b/.claude/rules/systems.md
new file mode 120000
index 00000000..b20dbdc3
--- /dev/null
+++ b/.claude/rules/systems.md
@@ -0,0 +1 @@
+../../.cursor/rules/systems.mdc
\ No newline at end of file
diff --git a/.claude/rules/tools.md b/.claude/rules/tools.md
new file mode 120000
index 00000000..9c7547e5
--- /dev/null
+++ b/.claude/rules/tools.md
@@ -0,0 +1 @@
+../../.cursor/rules/tools.mdc
\ No newline at end of file
diff --git a/.claude/rules/viewer-isolation.md b/.claude/rules/viewer-isolation.md
new file mode 120000
index 00000000..f484584b
--- /dev/null
+++ b/.claude/rules/viewer-isolation.md
@@ -0,0 +1 @@
+../../.cursor/rules/viewer-isolation.mdc
\ No newline at end of file
diff --git a/.cursor/rules/creating-rules.mdc b/.cursor/rules/creating-rules.mdc
new file mode 100644
index 00000000..4ac11580
--- /dev/null
+++ b/.cursor/rules/creating-rules.mdc
@@ -0,0 +1,78 @@
+---
+description: How to create and maintain project rules
+globs:
+alwaysApply: false
+---
+
+# Creating Rules
+
+Rules live in two places and are kept in sync via symlinks:
+
+- `.cursor/rules/<rule-name>.mdc` — source of truth (Cursor format)
+- `.claude/rules/<rule-name>.md` — symlink pointing to the cursor file
+
+## Workflow
+
+**1. Write the rule in `.cursor/rules/`**
+
+```
+.cursor/rules/my-rule.mdc
+```
+
+**2. Create a symlink in `.claude/rules/`**
+
+```bash
+ln -s ../../.cursor/rules/my-rule.mdc .claude/rules/my-rule.md
+```
+
+The `../../` prefix is required because the symlink lives two levels deep.
+
+**3. Verify**
+
+```bash
+ls -la .claude/rules/my-rule.md
+# → .claude/rules/my-rule.md -> ../../.cursor/rules/my-rule.mdc
+```
+
+## Rule File Format
+
+```markdown
+---
+description: One-line summary of what this rule covers
+globs:
+alwaysApply: false
+---
+
+# Rule Title
+
+Short intro paragraph.
+
+## Section
+
+Concrete guidance with examples.
+```
+
+- Set `alwaysApply: true` only for rules that apply to every file in the project.
+- Use `globs` to scope a rule to specific paths (e.g. `packages/viewer/**`).
+
+## Good Practices
+
+- Keep rules under 500 lines. Split large rules into smaller focused files.
+- Include concrete examples or reference real files with `@filename`.
+- Add a rule when the same mistake has been made more than once — not preemptively.
+- Prefer showing a correct example over listing prohibitions.
+
+## Existing Rules
+
+| Rule | Covers |
+|---|---|
+| `creating-rules` | This file — how to add rules |
+| `renderers` | Node renderer pattern in `packages/viewer` |
+| `systems` | Core and viewer systems architecture |
+| `tools` | Editor tools structure in `apps/editor` |
+| `viewer-isolation` | Keeping `@pascal-app/viewer` editor-agnostic |
+| `scene-registry` | Global node ID → Object3D map and `useRegistry` |
+| `selection-managers` | Two-layer selection (viewer + editor), events, outliner |
+| `events` | Typed event bus — emitting and listening to node and grid events |
+| `node-schemas` | Zod schema pattern for node types, createNode, updateNode |
+| `spatial-queries` | Placement validation (canPlaceOnFloor/Wall/Ceiling) for tools |
diff --git a/.cursor/rules/events.mdc b/.cursor/rules/events.mdc
new file mode 100644
index 00000000..ec751cfb
--- /dev/null
+++ b/.cursor/rules/events.mdc
@@ -0,0 +1,87 @@
+---
+description: Typed event bus — emitting and listening to node and grid events
+globs: packages/core/src/events/**,packages/viewer/**,apps/editor/**
+alwaysApply: false
+---
+
+# Events
+
+The event bus (`emitter`) is a global `mitt` instance typed with `EditorEvents`. It decouples renderers (which emit) from selection managers and tools (which listen).
+
+**Source**: @packages/core/src/events/bus.ts
+
+## Event Key Format
+
+```
+<nodeType>:<suffix>
+```
+
+Example keys: `wall:click`, `item:enter`, `door:double-click`, `grid:pointerdown`
+
+### Node Types
+`wall` `item` `site` `building` `level` `zone` `slab` `ceiling` `roof` `window` `door`
+
+### Suffixes
+```ts
+'click' | 'move' | 'enter' | 'leave' | 'pointerdown' | 'pointerup' | 'context-menu' | 'double-click'
+```
+
+The `grid:*` events fire when the user interacts with empty space (no node hit). They are **not** emitted by a mesh — `useGridEvents(gridY)` (@apps/editor/hooks/use-grid-events.ts) manually raycasts against a ground plane and calls `emitter.emit('grid:click', …)`. Mount it in any tool or editor component that needs empty-space interactions.
+
+## NodeEvent Shape
+
+```ts
+interface NodeEvent<T extends AnyNode = AnyNode> {
+  node: T                                  // typed node that triggered the event
+  position: [number, number, number]       // world-space hit position
+  localPosition: [number, number, number]  // object-local hit position
+  normal?: [number, number, number]        // face normal, if available
+  stopPropagation: () => void
+  nativeEvent: ThreeEvent<PointerEvent>
+}
+```
+
+Grid events only carry `position` and `nativeEvent` (no `node`).
+
+## Emitting
+
+Renderers emit via `useNodeEvents` — never call `emitter.emit` directly in a renderer:
+
+```tsx
+// packages/viewer/src/hooks/use-node-events.ts
+const events = useNodeEvents(node, 'wall')
+return <mesh ref={ref} {...events} />
+```
+
+`useNodeEvents` converts R3F `ThreeEvent` into a `NodeEvent` and emits `wall:click`, `wall:enter`, etc. It suppresses events while the camera is dragging.
+
+## Listening
+
+Listen in a `useEffect`. Always clean up with `emitter.off` using the **same function reference**:
+
+```ts
+// Single event
+useEffect(() => {
+  const handler = (e: WallEvent) => { /* … */ }
+  emitter.on('wall:click', handler)
+  return () => emitter.off('wall:click', handler)
+}, [])
+
+// Multiple node types, same handler
+useEffect(() => {
+  const types = ['wall', 'slab', 'door'] as const
+  const handler = (e: NodeEvent) => { /* … */ }
+  types.forEach(t => emitter.on(`${t}:click`, handler as any))
+  return () => types.forEach(t => emitter.off(`${t}:click`, handler as any))
+}, [])
+```
+
+See @apps/editor/components/editor/selection-manager.tsx for a full multi-type listener example.
+
+## Rules
+
+- **Renderers only emit, never listen.** Listening belongs in selection managers, tools, or systems.
+- **Always clean up.** Forgetting `emitter.off` causes duplicate handlers and memory leaks.
+- **Use the same function reference** for `on` and `off`. Anonymous functions inside `useEffect` are fine as long as the ref is captured in the same scope.
+- **Don't use emitter for state.** It's for one-shot interaction events. Persistent state goes in `useScene`, `useViewer`, or `useEditor`.
+- **`stopPropagation`** prevents the event from being handled by overlapping listeners (e.g. a door on a wall). Call it when a handler should be the final consumer.
diff --git a/.cursor/rules/node-schemas.mdc b/.cursor/rules/node-schemas.mdc
new file mode 100644
index 00000000..6cbc9080
--- /dev/null
+++ b/.cursor/rules/node-schemas.mdc
@@ -0,0 +1,94 @@
+---
+description: Node type definitions, Zod schema pattern, and how to create nodes in the scene
+globs: packages/core/src/schema/**
+alwaysApply: false
+---
+
+# Node Schemas
+
+All node types are defined as Zod schemas in `packages/core/src/schema/nodes/`. Each schema extends `BaseNode` and exports both the schema and its inferred TypeScript type.
+
+**Sources**: @packages/core/src/schema/base.ts, @packages/core/src/schema/nodes/
+
+## BaseNode
+
+Every node shares these fields:
+
+```ts
+{
+  object: 'node'            // always literal 'node'
+  id: string                // typed ID e.g. "wall_abc123"
+  type: string              // node type discriminator e.g. "wall"
+  name?: string             // optional display name
+  parentId: string | null   // parent node ID; null = root
+  visible: boolean          // defaults to true
+  metadata: Record<string, unknown>  // arbitrary JSON, defaults to {}
+}
+```
+
+## Defining a New Node Type
+
+```ts
+// packages/core/src/schema/nodes/my-node.ts
+import { z } from 'zod'
+import { BaseNode, objectId, nodeType } from '../base'
+
+export const MyNode = BaseNode.extend({
+  id: objectId('my-node'),      // generates IDs like "my-node_abc123"
+  type: nodeType('my-node'),    // sets literal type discriminator
+  // add node-specific fields:
+  width: z.number().default(1),
+  label: z.string().optional(),
+}).describe('My node — one-line description of what it represents')
+
+export type MyNode = z.infer<typeof MyNode>
+export type MyNodeId = MyNode['id']
+```
+
+Then add `MyNode` to the `AnyNode` union in `packages/core/src/schema/types.ts`.
+
+## Creating Nodes in Tools
+
+Always use `.parse()` to validate and generate a proper typed ID. Never construct a plain object manually.
+
+```ts
+import { WallNode } from '@pascal-app/core'
+import { useScene } from '@pascal-app/core'
+
+// 1. Parse validates and fills defaults (including auto-generated id)
+const wall = WallNode.parse({ name: 'Wall 1', start: [0, 0], end: [5, 0] })
+
+// 2. createNode(node, parentId?) inserts it into the scene
+const { createNode } = useScene.getState()
+createNode(wall, levelId)
+```
+
+For batch creation:
+
+```ts
+const { createNodes } = useScene.getState()
+createNodes([
+  { node: WallNode.parse({ start: [0, 0], end: [5, 0] }), parentId: levelId },
+  { node: WallNode.parse({ start: [5, 0], end: [5, 4] }), parentId: levelId },
+])
+```
+
+## Updating Nodes
+
+```ts
+const { updateNode } = useScene.getState()
+updateNode(wall.id, { height: 2.8 })   // partial update, merges with existing
+```
+
+## Real Examples
+
+- **Simple geometry node**: @packages/core/src/schema/nodes/wall.ts — `start`, `end`, `thickness`, `height`
+- **Polygon node**: @packages/core/src/schema/nodes/slab.ts — `polygon: [number, number][]`, `holes`
+- **Positioned node**: @packages/core/src/schema/nodes/item.ts — `position`, `rotation`, `scale`, `asset`
+
+## Rules
+
+- **Always use `.parse()`** — it generates the correct ID prefix and fills defaults. `WallNode.parse({...})` not `{ type: 'wall', id: '...' }`.
+- **Never hardcode IDs.** Let `objectId('type')` generate them.
+- **Add new node types to `AnyNode`** in `types.ts` or they won't be accepted by the store.
+- **Keep schemas in `packages/core`**, not in the viewer or editor — the schema is shared by all packages.
diff --git a/.cursor/rules/renderers.mdc b/.cursor/rules/renderers.mdc
new file mode 100644
index 00000000..09acabd9
--- /dev/null
+++ b/.cursor/rules/renderers.mdc
@@ -0,0 +1,68 @@
+---
+description: Node renderer pattern in packages/viewer
+globs: packages/viewer/**
+alwaysApply: false
+---
+
+# Renderers
+
+Renderers live in `packages/viewer/src/components/renderers/`. Each renderer is responsible for one node type's Three.js geometry and materials — nothing else.
+
+## Dispatch Chain
+
+```
+<SceneRenderer>          — iterates rootNodeIds from useScene
+  └─ <NodeRenderer>      — switches on node.type, renders the matching component
+       └─ <WallRenderer> — (or SlabRenderer, DoorRenderer, …)
+```
+
+See @packages/viewer/src/components/renderers/scene-renderer.tsx and @packages/viewer/src/components/renderers/node-renderer.tsx.
+
+## Renderer Responsibilities
+
+A renderer **should**:
+- Read its node from `useScene` via the node's ID
+- Register its mesh(es) with `useRegistry()` so other systems can look them up
+- Subscribe to pointer events via `useNodeEvents()`
+- Render geometry and apply materials based on node properties
+
+A renderer **must not**:
+- Run geometry generation logic (that belongs in a System)
+- Import anything from `apps/editor`
+- Manage selection state directly (use `useViewer` for read, emit events for write)
+- Perform expensive per-frame calculations in the component body
+
+## Example — Minimal Renderer
+
+```tsx
+// packages/viewer/src/components/renderers/my-node/index.tsx
+import { useRegistry } from '@pascal-app/core'
+import { useNodeEvents } from '../../hooks/use-node-events'
+import { useScene } from '@pascal-app/core'
+
+export function MyNodeRenderer({ node }: { node: MyNode }) {
+  const ref = useRef<Mesh>(null!)
+  useRegistry(node.id, 'my-node', ref)   // 3 args: id, type, ref — no return value
+  const events = useNodeEvents(node, 'my-node')
+
+  return (
+    <mesh ref={ref} {...events}>
+      <boxGeometry args={[node.width, node.height, node.depth]} />
+      <meshStandardMaterial color={node.color} />
+    </mesh>
+  )
+}
+```
+
+## Adding a New Node Type
+
+1. Create `packages/viewer/src/components/renderers/<type>/index.tsx`
+2. Add a case to `NodeRenderer` in `node-renderer.tsx`
+3. Add the corresponding system in `packages/core/src/systems/` if the node needs derived geometry
+4. Export from `packages/viewer/src/index.ts` if needed externally
+
+## Performance Notes
+
+- Use `useMemo` for geometry that depends on node properties — avoid recreating on every render.
+- For complex cutout or boolean geometry, delegate to a System (e.g. `WallCutout`).
+- Register one mesh per node ID; if a renderer spawns multiple meshes, use a group ref or pick the primary one for registry.
diff --git a/.cursor/rules/scene-registry.mdc b/.cursor/rules/scene-registry.mdc
new file mode 100644
index 00000000..e7e63d44
--- /dev/null
+++ b/.cursor/rules/scene-registry.mdc
@@ -0,0 +1,80 @@
+---
+description: Scene registry pattern — mapping node IDs to live THREE.Object3D instances
+globs: packages/core/src/hooks/scene-registry/**,packages/viewer/**
+alwaysApply: false
+---
+
+# Scene Registry
+
+The scene registry is a global, mutable map that links node IDs to their live `THREE.Object3D` instances. It avoids tree traversal and lets systems and selection managers do O(1) lookups.
+
+**Source**: @packages/core/src/hooks/scene-registry/scene-registry.ts
+
+## Structure
+
+```ts
+export const sceneRegistry = {
+  nodes: new Map<string, THREE.Object3D>(),   // id → Object3D
+  byType: {
+    wall: new Set<string>(),
+    slab: new Set<string>(),
+    item: new Set<string>(),
+    // … one Set per node type
+  },
+}
+```
+
+`nodes` is the primary lookup. `byType` lets systems iterate all objects of one type without scanning the whole map.
+
+## Registering in a Renderer
+
+Every renderer must call `useRegistry` with a `ref` to its root mesh or group. Registration is synchronous (`useLayoutEffect`) so it's available before the first paint.
+
+```tsx
+import { useRegistry } from '@pascal-app/core'
+
+export function WallRenderer({ node }: { node: WallNode }) {
+  const ref = useRef<Mesh>(null!)
+  useRegistry(node.id, 'wall', ref)   // ← required in every renderer
+
+  return <mesh ref={ref} … />
+}
+```
+
+The hook handles both registration on mount and cleanup on unmount automatically.
+
+## Looking Up Objects
+
+Anywhere outside the renderer — in systems, selection managers, export logic:
+
+```ts
+// Single lookup
+const obj = sceneRegistry.nodes.get(nodeId)
+if (obj) { /* use obj */ }
+
+// Iterate all walls
+for (const id of sceneRegistry.byType.wall) {
+  const obj = sceneRegistry.nodes.get(id)
+}
+```
+
+## Rules
+
+- **One registration per node ID.** If a renderer spawns multiple meshes, register the outermost group (the one that represents the node).
+- **Never hold a stale reference.** Always read from `sceneRegistry.nodes.get(id)` at the time you need it — don't cache the result across frames.
+- **Don't mutate the registry manually.** Only `useRegistry` should add/remove entries. Systems and selection managers are read-only consumers.
+- **Core systems must not use the registry.** They work with plain node data. Only viewer systems and selection managers may do Three.js object lookups.
+
+## Outliner Sync
+
+The `outliner` in `useViewer` holds live `Object3D[]` arrays used by the post-processing outline pass. Selection managers sync them imperatively for performance (array mutation rather than new allocations):
+
+```ts
+outliner.selectedObjects.length = 0
+for (const id of selection.selectedIds) {
+  const obj = sceneRegistry.nodes.get(id)
+  if (obj) outliner.selectedObjects.push(obj)
+}
+```
+
+See @packages/viewer/src/components/viewer/selection-manager.tsx for the full sync pattern.
diff --git a/.cursor/rules/selection-managers.mdc b/.cursor/rules/selection-managers.mdc
new file mode 100644
index 00000000..3a83c094
--- /dev/null
+++ b/.cursor/rules/selection-managers.mdc
@@ -0,0 +1,95 @@
+---
+description: Selection managers — two-layer architecture for viewer and editor selection
+globs: packages/viewer/src/components/viewer/selection-manager.tsx,apps/editor/components/editor/selection-manager.tsx
+alwaysApply: false
+---
+
+# Selection Managers
+
+There are two selection managers. They are separate components, not the same component configured differently.
+
+| Component | Location | Knows about |
+|---|---|---|
+| `SelectionManager` | `packages/viewer/src/components/viewer/selection-manager.tsx` | Viewer state only |
+| `SelectionManager` (editor) | `apps/editor/components/editor/selection-manager.tsx` | Phase, mode, tool state |
+
+The viewer's manager is the default. The editor mounts its own manager as a child of `<Viewer>`, overriding the default behaviour via the viewer-isolation pattern.
+
+---
+
+## How Selection Works
+
+**Event flow:**
+
+```
+useNodeEvents(node, type) on a renderer mesh
+  → emitter.emit('wall:click', NodeEvent)
+  → SelectionManager listens via emitter.on(…)
+  → calls useViewer.setSelection(…)
+  → outliner sync re-runs → Three.js outline updates
+```
+
+`useNodeEvents` returns R3F pointer handlers. Spread them onto the mesh:
+
+```tsx
+const events = useNodeEvents(node, 'wall')
+return <mesh ref={ref} {...events} />
+```
+
+Events are suppressed during camera drag (`useViewer.getState().cameraDragging`).
+
+---
+
+## Viewer Selection Manager
+
+Hierarchical path: **Building → Level → Zone → Elements**
+
+At each level, only the next tier is selectable. Clicking outside deselects. The path is stored in `useViewer`:
+
+```ts
+type SelectionPath = {
+  buildingId: string | null
+  levelId: string | null
+  zoneId: string | null
+  selectedIds: string[]   // walls, items, slabs, etc.
+}
+```
+
+`setSelection` has a hierarchy guard: setting `levelId` without `buildingId` resets children. Use `resetSelection()` to clear everything.
+
+Multi-select: `Ctrl/Meta + click` toggles an ID in `selectedIds`. Regular click replaces it.
+
+---
+
+## Editor Selection Manager
+
+Extends selection with phase awareness from `useEditor`. The viewer's `SelectionManager` is **not** mounted in the editor; this one takes its place (injected as a child of `<Viewer>`).
+
+```
+phase: 'site'      → selectable: buildings
+phase: 'structure' → selectable: walls, zones, slabs, ceilings, roofs, doors, windows
+  structureLayer: 'zones'    → only zones
+  structureLayer: 'elements' → all structure types
+phase: 'furnish'   → selectable: furniture items only
+```
+
+Clicking a node of a different phase auto-switches the phase. Double-click drills into a context level.
+
+---
+
+## Rules
+
+- **Never add selection logic to renderers.** Renderers spread `useNodeEvents` events and stop there. All selection decisions live in the selection manager.
+- **Never add editor phase logic to the viewer's SelectionManager.** Phase, mode, and tool awareness belong exclusively in the editor's selection manager.
+- **`useViewer` is the single source of truth for selection state.** Both managers read and write through `setSelection` / `resetSelection`. Nothing else should mutate `selection` directly.
+- **Outliner arrays are mutated in-place** (not replaced) for performance. Don't assign new arrays to `outliner.selectedObjects` or `outliner.hoveredObjects`.
+- **Hover is a separate scalar** (`hoveredId: string | null`), not part of `selectedIds`. Update it via `setHoveredId`.
+
+---
+
+## Adding Selectability to a New Node Type
+
+1. Add the type to `SelectableNodeType` in the viewer store / selection manager.
+2. Make sure its renderer calls `useNodeEvents(node, type)` and spreads the handlers.
+3. Add a case to whichever selection strategy needs it (viewer hierarchy level or editor phase).
+4. Ensure `useRegistry` is called in the renderer so the outliner can highlight it.
diff --git a/.cursor/rules/spatial-queries.mdc b/.cursor/rules/spatial-queries.mdc
new file mode 100644
index 00000000..b33ae5f3
--- /dev/null
+++ b/.cursor/rules/spatial-queries.mdc
@@ -0,0 +1,106 @@
+---
+description: Placement validation for tools — canPlaceOnFloor, canPlaceOnWall, canPlaceOnCeiling
+globs: apps/editor/components/tools/**
+alwaysApply: false
+---
+
+# Spatial Queries
+
+`useSpatialQuery()` validates whether an item can be placed at a given position without overlapping existing items. Every placement tool must call it before committing a node to the scene.
+
+**Source**: @packages/core/src/hooks/spatial-grid/use-spatial-query.ts
+
+## Hook
+
+```ts
+const { canPlaceOnFloor, canPlaceOnWall, canPlaceOnCeiling } = useSpatialQuery()
+```
+
+All three methods return `{ valid: boolean; conflictIds: string[] }`.
+`canPlaceOnWall` additionally returns `adjustedY: number` (snapped height).
+
+---
+
+## canPlaceOnFloor
+
+```ts
+canPlaceOnFloor(
+  levelId: string,
+  position: [number, number, number],
+  dimensions: [number, number, number],   // scaled width/height/depth
+  rotation: [number, number, number],
+  ignoreIds?: string[],                   // pass [draftItem.id] to exclude self
+): { valid: boolean; conflictIds: string[] }
+```
+
+**Usage in a tool:**
+```ts
+const pos: [number, number, number] = [x, 0, z]
+const { valid } = canPlaceOnFloor(levelId, pos, getScaledDimensions(item), item.rotation, [item.id])
+if (valid) createNode(item, levelId)
+```
+
+---
+
+## canPlaceOnWall
+
+```ts
+canPlaceOnWall(
+  levelId: string,
+  wallId: string,
+  localX: number,          // distance along wall from start
+  localY: number,          // height from floor
+  dimensions: [number, number, number],
+  attachType: 'wall' | 'wall-side',  // 'wall' needs clearance both sides; 'wall-side' only one
+  side?: 'front' | 'back',
+  ignoreIds?: string[],
+): { valid: boolean; conflictIds: string[]; adjustedY: number }
+```
+
+`adjustedY` contains the snapped Y so items sit flush on the slab — always use it instead of the raw `localY`:
+
+```ts
+const { valid, adjustedY } = canPlaceOnWall(levelId, wallId, x, y, dims, 'wall', undefined, [item.id])
+if (valid) updateNode(item.id, { wallT: x, wallY: adjustedY })
+```
+
+---
+
+## canPlaceOnCeiling
+
+```ts
+canPlaceOnCeiling(
+  ceilingId: string,
+  position: [number, number, number],
+  dimensions: [number, number, number],
+  rotation: [number, number, number],
+  ignoreIds?: string[],
+): { valid: boolean; conflictIds: string[] }
+```
+
+---
+
+## Slab Elevation
+
+When items rest on a slab (not flat ground), use these to get the correct Y:
+
+```ts
+import { spatialGridManager } from '@pascal-app/core'
+
+// Y at a single point
+const y = spatialGridManager.getSlabElevationAt(levelId, x, z)
+
+// Y considering the item's full footprint (highest slab point under item)
+const y = spatialGridManager.getSlabElevationForItem(levelId, position, dimensions, rotation)
+```
+
+---
+
+## Rules
+
+- **Always pass `[item.id]` in `ignoreIds`** when validating a draft item that already exists in the scene — otherwise it collides with itself.
+- **Use `adjustedY` from `canPlaceOnWall`** — don't use the raw cursor Y for wall-mounted items.
+- **Use `getScaledDimensions(item)`** (@packages/core/src/schema/nodes/item.ts) to account for item scale, not the raw `asset.dimensions`.
+- Validate on every pointer move for live feedback (highlight ghost red/green). Only `createNode` / `updateNode` on pointer up or click.
+
+See @apps/editor/components/tools/item/use-placement-coordinator.tsx for a full implementation.
diff --git a/.cursor/rules/systems.mdc b/.cursor/rules/systems.mdc
new file mode 100644
index 00000000..c35a4a9f
--- /dev/null
+++ b/.cursor/rules/systems.mdc
@@ -0,0 +1,92 @@
+---
+description: Core and viewer systems architecture
+globs: packages/core/src/systems/**,packages/viewer/src/systems/**
+alwaysApply: false
+---
+
+# Systems
+
+Systems own business logic, geometry generation, and constraints. They run in the Three.js frame loop and are never rendered directly.
+
+## Two Kinds of Systems
+
+### Core Systems — `packages/core/src/systems/`
+
+Pure logic: no rendering, no Three.js objects. They read nodes from `useScene`, compute derived values (geometry, constraints), and write results back.
+
+| System | Responsibility |
+|---|---|
+| `WallSystem` | Wall mitering, corner joints |
+| `SlabSystem` | Polygon-based floor/roof generation |
+| `CeilingSystem` | Polygon-based ceiling generation |
+| `RoofSystem` | Pitched roof shape |
+| `DoorSystem` | Placement constraints on walls |
+| `WindowSystem` | Placement constraints on walls |
+| `ItemSystem` | Item transforms, collision |
+
+### Viewer Systems — `packages/viewer/src/systems/`
+
+Access Three.js objects (via `useRegistry`) and manage rendering side-effects.
+
+| System | Responsibility |
+|---|---|
+| `LevelSystem` | Stacked / exploded / solo / manual level positions |
+| `WallCutout` | Cuts door/window holes in wall geometry |
+| `ZoneSystem` | Zone display and label placement |
+| `InteractiveSystem` | Item toggles and sliders in the scene |
+| `GuideSystem` | Temporary helper geometry |
+| `ScanSystem` | Point cloud rendering |
+
+## Pattern
+
+Systems are React components that render nothing (`return null`) and use `useFrame` for per-frame logic.
+
+```tsx
+// packages/core/src/systems/my-system.tsx
+import { useFrame } from '@react-three/fiber'
+import { useScene } from '../store/use-scene'
+
+export function MySystem() {
+  const nodes = useScene(s => s.nodes)
+
+  useFrame(() => {
+    // compute and write back derived state
+  })
+
+  return null
+}
+```
+
+Core and viewer systems are mounted inside `<Viewer>` alongside renderers. See @packages/viewer/src/components/viewer/index.tsx for the mount order.
+
+**Systems are a customization point.** Any consumer of `<Viewer>` — the editor app, an embed, a read-only preview — can inject its own systems as children. This is how editor-specific behaviour (space detection, tool feedback) is added without touching the viewer package.
+
+## Rules
+
+- **Core systems must not import Three.js** — they work with plain data.
+- **Viewer systems must not contain business logic** — delegate to core if the rule is domain-level.
+- **Never duplicate logic** between a system and a renderer — if the renderer needs it, the system should compute and store it, and the renderer reads the result.
+- Systems should be **idempotent**: given the same nodes, they produce the same output.
+- Mark nodes as `dirty` in the scene store to signal that a system should re-run. Avoid running expensive logic every frame without a dirty check.
+
+## Adding a New System
+
+1. Decide the scope:
+   - **Domain logic** → `packages/core/src/systems/`
+   - **Viewer rendering side-effect** → `packages/viewer/src/systems/` — mount in `packages/viewer/src/components/viewer/index.tsx`
+   - **Editor-specific or integration-specific** → keep it in the consuming app (e.g. `apps/editor/components/systems/`) and inject it as a child of `<Viewer>`
+
+2. Create `<name>-system.tsx` in the appropriate directory.
+
+3. Mount it in the right place:
+   - Viewer-internal systems go in `packages/viewer/src/components/viewer/index.tsx`
+   - App-specific systems are injected as children from outside:
+     ```tsx
+     // apps/editor — editor injects its own systems without modifying the viewer
+     <Viewer>
+       <MyEditorSystem />
+       <ToolManager />
+     </Viewer>
+     ```
+
+4. **Mount order matters.** Most viewer systems run *after* renderers in the JSX tree — they consume `sceneRegistry` data that renderers populate on mount. Only place a system before renderers if it explicitly does not read the registry.
diff --git a/.cursor/rules/tools.mdc b/.cursor/rules/tools.mdc
new file mode 100644
index 00000000..840e1d57
--- /dev/null
+++ b/.cursor/rules/tools.mdc
@@ -0,0 +1,75 @@
+---
+description: Editor tools structure in apps/editor
+globs: apps/editor/components/tools/**
+alwaysApply: false
+---
+
+# Tools
+
+Tools are React components that capture user input (pointer, keyboard) and translate it into `useScene` mutations. They live exclusively in `apps/editor/components/tools/`.
+
+## Lifecycle
+
+`ToolManager` reads `useEditor` (phase + mode + tool) and mounts the active tool component. When the tool changes, the old component unmounts, cleaning up any transient state.
+
+See @apps/editor/components/tools/tool-manager.tsx.
+
+## Tool Categories by Phase
+
+**Site**
+- `site-boundary-editor` — draw/edit property boundary polygon
+
+**Structure**
+- `wall-tool` — draw walls segment by segment
+- `slab-tool` + `slab-boundary-editor` + `slab-hole-editor`
+- `ceiling-tool` + `ceiling-boundary-editor` + `ceiling-hole-editor`
+- `roof-tool`
+- `door-tool` + `door-move-tool`
+- `window-tool` + `window-move-tool`
+- `item-tool` + `item-move-tool`
+- `zone-tool` + `zone-boundary-editor`
+
+**Furnish**
+- `item-tool` — place furniture
+
+**Shared utilities**
+- `polygon-editor` — reusable boundary/hole editing logic
+- `cursor-sphere` — 3D cursor visualisation
+
+## Pattern
+
+```tsx
+// apps/editor/components/tools/my-tool/index.tsx
+import { useScene } from '@pascal-app/core'
+import { useEditor } from '../../store/use-editor'
+
+export function MyTool() {
+  const createNode = useScene(s => s.createNode)
+  const setTool = useEditor(s => s.setTool)
+
+  // Pointer handlers mutate the scene store directly.
+  // No local geometry — use a renderer for any preview mesh.
+
+  return (
+    <mesh onPointerDown={handleDown} onPointerMove={handleMove}>
+      {/* ghost / preview geometry only */}
+    </mesh>
+  )
+}
+```
+
+## Rules
+
+- **Tools only mutate `useScene`** — they do not call Three.js APIs directly.
+- **No business logic in tools** — delegate geometry/constraint rules to core systems.
+- **Preview geometry is local** — transient meshes shown while a tool is active live in the tool component, not in the scene store.
+- **Clean up on unmount** — remove any pending/incomplete nodes when the tool unmounts.
+- **Tools must not import from `@pascal-app/viewer`** — use the scene store and core hooks only.
+- Each tool should handle a single, well-scoped interaction. Split complex tools (e.g. "draw + move") into separate components selected by `useEditor`.
+
+## Adding a New Tool
+
+1. Create `apps/editor/components/tools/<name>/index.tsx`.
+2. Register the tool in `ToolManager` under the correct phase and mode.
+3. Add the tool identifier to the `useEditor` tool union type.
+4. If the tool requires new node types, add schema + renderer + system first.
diff --git a/.cursor/rules/viewer-isolation.mdc b/.cursor/rules/viewer-isolation.mdc
new file mode 100644
index 00000000..a715747b
--- /dev/null
+++ b/.cursor/rules/viewer-isolation.mdc
@@ -0,0 +1,88 @@
+---
+description: Viewer must be editor-agnostic — controlled from outside via props and children
+globs: packages/viewer/**
+alwaysApply: false
+---
+
+# Viewer Isolation
+
+`@pascal-app/viewer` is a standalone 3D canvas library. It must never know about editor-specific features, UI state, or tools. This keeps it usable in the read-only `/viewer/[id]` route and in any future embedding context.
+
+## The Rule
+
+> The viewer is controlled from outside. It exposes control points (props, callbacks, children). It never reaches into `apps/editor`.
+
+## Forbidden in `packages/viewer`
+
+```ts
+// ❌ Never import from the editor app
+import { useEditor } from '@/store/use-editor'
+import { ToolManager } from '@/components/tools/tool-manager'
+
+// ❌ Never reference editor-specific concepts
+if (isEditorMode) { … }
+```
+
+## Correct Pattern — Pass Control from Outside
+
+The editor mounts the viewer and passes what it needs:
+
+```tsx
+// apps/editor/components/editor-canvas.tsx  ✅
+import { Viewer } from '@pascal-app/viewer'
+import { ToolManager } from '../tools/tool-manager'
+import { useEditor } from '../../store/use-editor'
+
+export function EditorCanvas() {
+  const { selection } = useViewer()
+
+  return (
+    <Viewer
+      theme="light"
+      onSelect={(id) => useViewer.getState().setSelection(id)}
+      onExport={handleExport}
+    >
+      {/* Editor injects tools as children — viewer renders them inside the canvas */}
+      <ToolManager />
+    </Viewer>
+  )
+}
+```
+
+The viewer accepts `children` and renders them inside the R3F canvas. This is the extension point for tools, overlays, and editor-specific systems.
+
+## Viewer's Own State (`useViewer`)
+
+The viewer store contains **only presentation state**:
+
+- `selection` — which nodes are highlighted
+- `cameraMode` — perspective / orthographic
+- `levelMode` — stacked / exploded / solo / manual
+- `wallMode` — up / cutaway / down
+- `theme` — light / dark
+- Display toggles: `showScans`, `showGuides`, `showGrid`
+
+If a piece of state is only meaningful inside the editor (e.g. active tool, phase, edit mode) — it belongs in `useEditor`, not `useViewer`.
+
+## Nested Viewer for Editor-Specific Features
+
+When an editor feature needs to live "inside" the canvas but must not pollute the viewer package, inject it as a child:
+
+```tsx
+// ✅ Editor-specific overlay injected as child
+<Viewer>
+  <SelectionBoxOverlay />   {/* editor only */}
+  <SnapIndicator />         {/* editor only */}
+  <ToolManager />           {/* editor only */}
+</Viewer>
+```
+
+This pattern lets the viewer stay ignorant of these components while they still have access to the R3F context.
+
+## Checklist Before Adding Code to `packages/viewer`
+
+- [ ] Does this feature make sense in the read-only viewer route?
+- [ ] Does it reference `useEditor`, tool state, or phase/mode?
+- [ ] Could it be passed in as a prop or child instead?
+
+If any answer is "editor-specific", keep it in `apps/editor` and inject it via children or props.
diff --git a/AGENTS.md b/AGENTS.md
new file mode 100644
index 00000000..e608940a
--- /dev/null
+++ b/AGENTS.md
@@ -0,0 +1,95 @@
+# Pascal Editor V2 — Architecture
+
+## Project Structure
+
+Monorepo managed with Turborepo. Packages are shared libraries; apps are deployable applications.
+
+```
+apps/
+  editor/          # Main Next.js app (editor + public routes)
+packages/
+  core/            # Scene schema, state, systems, spatial logic
+  viewer/          # 3D canvas component (React Three Fiber)
+  auth/            # Authentication (better-auth + Supabase)
+  db/              # Database layer (Drizzle ORM + Supabase)
+  ui/              # Shared React UI components
+```
+
+---
+
+## packages/core
+
+Central library — no UI, no rendering. Everything else depends on it.
+
+- **schema/** — TypeScript types for all node types (`Wall`, `Slab`, `Door`, `Item`, etc.)
+- **store/** — Zustand scene store (`useScene`) with undo/redo via Zundo
+- **systems/** — Per-element business logic: geometry generation, constraints (`WallSystem`, `SlabSystem`, `DoorSystem`, …)
+- **events/** — Typed event bus for node changes
+- **hooks/** — `useRegistry` (node ID → THREE.Object3D), `useSpatialGrid` (2D spatial index)
+- **lib/** — Space detection, asset storage, polygon utilities
+
+Node storage is a flat dictionary (`nodes: Record<id, AnyNode>`). Systems are pure logic that runs in the render loop; they read nodes and write back derived geometry.
+
+---
+
+## packages/viewer
+
+3D canvas component — presentation only, no editor concerns.
+
+- **components/viewer/** — Root `<Viewer>` canvas, camera, lights, post-processing, selection manager
+- **components/renderers/** — One renderer per node type (`WallRenderer`, `SlabRenderer`, …), dispatched by `NodeRenderer` → `SceneRenderer`
+- **systems/** — Viewer-specific systems: `LevelSystem` (stacked/exploded/solo), `WallCutout`, `ZoneSystem`, `InteractiveSystem`
+- **store/** — `useViewer`: selection path, camera mode, level mode, wall mode, theme, display toggles
+
+The viewer accepts external props and callbacks (`onSelect`, `onExport`, children) to expose control points. It must not import anything from `apps/editor`.
+
+---
+
+## apps/editor
+
+Next.js 16 app. Composes `@pascal-app/viewer` and `@pascal-app/core` into a full editing experience.
+
+- **app/editor/[projectId]/** — Main editor route
+- **app/viewer/[id]/** — Read-only preview route
+- **store/use-editor.tsx** — `useEditor`: phase (`site | structure | furnish`), mode (`select | edit | delete | build`), active tool
+- **components/tools/** — One component per tool, coordinated by `ToolManager`
+- **components/systems/** — Editor-side systems that integrate with viewer (e.g. space detection for cutaway)
+- **components/editor/** — Camera controls, export, menus, panels
+
+---
+
+## Data Flow
+
+```
+User input (pointer/keyboard)
+  → Tool component (apps/editor/components/tools/)
+  → useScene mutations
+  → Core systems recompute geometry
+  → Renderers re-render THREE meshes
+  → useViewer updates selection/hover
+```
+
+---
+
+## Key Conventions
+
+- **Flat nodes** — All scene nodes live in a single flat record; hierarchy is expressed via `parentId`.
+- **System/renderer split** — Systems own logic; renderers own geometry and material. Never mix.
+- **Viewer isolation** — `@pascal-app/viewer` must never import from `apps/editor`. Editor-specific behaviour (tools, systems, selection) is injected as children or props.
+- **Registry pattern** — `useRegistry()` maps node IDs to live THREE objects without tree traversal.
+- **Spatial grid** — 2D grid for fast wall/zone neighbourhood queries; avoid brute-force iteration.
+- **Node creation** — Always use `NodeType.parse({…})` then `createNode(node, parentId)`. Never construct raw node objects.
+
+---
+
+## Tech Stack
+
+| Layer | Technology |
+|---|---|
+| 3D | Three.js (WebGPU), React Three Fiber |
+| Framework | Next.js 16, React 19 |
+| State | Zustand + Zundo |
+| UI | Radix UI, Tailwind CSS 4 |
+| Database | Supabase PostgreSQL + Drizzle ORM |
+| Auth | better-auth |
+| Tooling | Biome, TypeScript 5.9, Turborepo |
diff --git a/CLAUDE.md b/CLAUDE.md
new file mode 120000
index 00000000..47dc3e3d
--- /dev/null
+++ b/CLAUDE.md
@@ -0,0 +1 @@
+AGENTS.md
\ No newline at end of file