Scene Composition logo

Scene Composition

Organization
SummerEngine
scene-composition

Scene structure: node hierarchy, when to extract sub-scenes, reusable prefabs, instance vs add node.

Overview

PublisherSummerEngine
Repositorysummer
Skill namescene-composition
Stars
91
Forks
8
Bundled files
1
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by SummerEngine on GitHub. Read the source before you install it.

Installation

Install the Scene Composition AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/SummerEngine/summer/tree/main/library/skills/scene-composition
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/SummerEngine/summer.git /tmp/summer
mkdir -p .claude/skills
cp -r /tmp/summer/library/skills/scene-composition .claude/skills/scene-composition
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Scene Composition in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Scene Composition on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Scene Composition is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Scene Composition for Summer Engine

Organize scenes for clarity, reuse, and MCP compatibility. Follow these conventions when building levels, characters, and UI.

Node Hierarchy Conventions

3D Scenes

World (Node3D)                    # Root or main container
├── Camera3D                      # Main camera
├── DirectionalLight3D            # Sun / primary light
├── WorldEnvironment              # Sky, ambient, fog
├── Level (Node3D)                # Level geometry
│   ├── Ground
│   ├── Walls
│   └── Platforms
├── Props (Node3D)                # Placed objects (trees, crates)
├── Enemies (Node3D)              # Enemy instances
└── Player                        # Player instance (or instantiated)

2D Scenes

Level (Node2D)
├── TileMapLayer                  # Or TileMap
├── Characters
├── Props
└── Effects

UI Scenes

CanvasLayer or Control
├── MarginContainer
│   ├── VBoxContainer
│   │   ├── Label
│   │   ├── Button
│   │   └── Button

Parent Paths for MCP

Use ./ for paths relative to the scene root:

PathMeaning
./Scene root
./WorldChild named "World" of root
./World/PlayerPlayer under World
./World/Props/Tree1Tree1 under Props under World

Never use /World (absolute) or World (missing ./). The engine expects ./ prefix.

When to Use Sub-Scenes

Use sub-scenes (separate .tscn files) when:

  • The same setup appears in multiple places (player, enemy, pickup)
  • The setup has many nodes and would clutter the main scene
  • You want to edit a prefab in isolation

Use inline nodes when:

  • The node is unique to this scene (main camera, level-specific light)
  • It's a simple one-off (single MeshInstance3D)

Creating a Sub-Scene

  1. Build the hierarchy in the main scene
  2. Select the root of what you want to extract
  3. Save as scene: summer_save_scene(scenePath="res://scenes/source.tscn", path="res://scenes/player.tscn"). scenePath names the scene being saved and is required; path is the optional save-as target. There is no "current scene" default.
  4. In the main scene, add it with summer_instantiate_scene(scenePath="res://main.tscn", parent="./World", scene="res://scenes/player.tscn", name="Player")

Practical approach: Create reusable scenes (player.tscn, enemy.tscn) as separate scene files, then instantiate them into levels.

InstantiateScene vs AddNode

UseToolExample
Built-in mesh (Box, Sphere)AddNode + SetPropsummer_add_node type=MeshInstance3D, then summer_set_prop mesh=BoxMesh
Existing .tscn prefabInstantiateScenesummer_instantiate_scene scene=res://player.tscn
Imported .glb modelImportFromUrl then InstantiateSceneImport first, then instantiate

Do not use summer_set_prop with mesh for a .glb path. Use summer_instantiate_scene for .tscn and .glb files.

Save Conventions

  • The mutation tools (summer_add_node, summer_set_prop, ...) already append one save per call, so a standalone summer_save_scene is only for a save-as or an explicit extra flush
  • Save-as: summer_save_scene(scenePath="res://main.tscn", path="res://scenes/level1.tscn")
  • Plain save: summer_save_scene(scenePath="res://main.tscn") — scenePath is required in both forms

Spatial Placement

When a 3D prop, character, collider, platform, wall-mounted object, shelf item, or tight-space asset must sit flush, avoid overlap, or preserve clearance, load the spatial-placement skill. Follow its inspect -> place -> Starcast -> correct -> verify loop instead of judging precise placement from transforms or a screenshot alone.

Common Mistakes

  1. Wrong parent path: ./NonExistent fails. Ensure the parent exists before adding children.
  2. Unnamed scene: Saving without a path fails if the scene was never saved. Use path for new scenes.
  3. Duplicate names: the engine auto-renames (Node, Node2, etc.). Use descriptive unique names.
  4. Mixing 2D and 3D: Don't put Node2D under Node3D or vice versa in the same hierarchy.

Fallback

MCP is the preferred path, not a requirement. Hand-written .tscn text loads and instantiates fine — including [ext_resource] lines with no uid= field, which the engine resolves by path (verified by loading such a scene on the shipped 4.6.1 build). Format version 3 is current.

What actually goes wrong when you hand-write scenes is mundane and checkable: a SubResource("id") with no matching [sub_resource id="id"] block, a parent= path naming a node that isn't declared above it, or an [ext_resource path=...] pointing at a file that hasn't been imported yet (--headless --import, or summer_import_from_url, before it will load).

So: use MCP when it's connected because the ops are undoable as one step and the engine validates paths for you. When it isn't, write the .tscn and then open it once to confirm it loads. Do not tell the user it's impossible.

Trap — Cross-scene transform leak when copying a scene as a template

When you duplicate a scene to spawn a sibling level (for example, level_1.tscn → level_2.tscn), every child instance carries its transform offset baked at the original level's coordinate system. If the source level places its player at world (223, 108, -1833) and instances a kill-volume scene at offset (512, -3, 512), those numbers are anchored to the source level's terrain origin. Drop the same kill-volume instance into a level whose player spawns at world (0, 1, 0) and the volume's wall colliders may end up slicing the new arena at world x = 0. The user sees a "glitching invisible line through the middle of the map" they can't pass through. The colliders are invisible under the floor, so visual diagnosis is hard. Only collision-walking through them reveals the problem.

Rule when copying a scene as a template:

  1. Audit every child instance for hardcoded transform = Transform3D(...) lines.
  2. Decide for each: keep, re-zero (origin = (0,0,0)), or drop entirely.
  3. The bigger the original level's world-space coordinates, the more dangerous the copy. Re-zero by default; you can move things back if needed.
  4. Levels that spawn the player at world origin (for example flat arenas centered on (0,0,0)) should drop ALL transform overrides from carry-over instances. Only re-add transforms relative to the new origin.

Trap — Third-person camera collision when foliage shouldn't block

SpringArm3D shortens the camera boom when it raycasts into geometry on the configured collision_mask. If your trees, bushes, or grass have collision so the player physically bumps them, on the same layer the SpringArm queries, the camera shortens for foliage too. Players see the camera awkwardly snap forward whenever they walk near a tree.

The pattern that works:

  1. Foliage gets its own physics layer. Conventionally layer 9 ("Foliage").
    • Foliage StaticBody3D: collision_layer = 256 (which is 2^8, layer 9).
  2. Player still bumps into foliage by including that layer in its collision_mask.
    • Player CharacterBody3D: collision_mask = 259 (which is 1 + 2 + 256, Entities + Level + Foliage).
  3. SpringArm uses a mask that catches solid level geometry but excludes foliage.
    • SpringArm: collision_mask = 3 (which is 1 + 2, Entities + Level only).
  4. For the visual fade the user expects when the camera is "behind" a tree, use a camera-occlusion fade instead of collision: if the project has one, add the foliage mesh to its group (for example camera_fade) so it ghosts the mesh when it crosses the camera-to-player segment.

Why per-layer split: there's no clean way to exclude a single body type from SpringArm3D's collision check via add_excluded_object for dynamically-spawned foliage. You'd have to register every tree at spawn time and re-register on level change. Layer filtering needs no upkeep.

Don't skip the player mask update. If you only move foliage to layer 9 without adding it to the player's mask, the player walks straight through trees.

Collaborative protocol

This skill creates and mutates scene files. Always ask before applying: "May I create res://scenes/player.tscn and instantiate it under ./World?".

From a shell

No Summer MCP tools in this session? Run each one as a command; when summer --version is missing or below 4.0, use npx -y summer-engine@latest in place of summer. Fields are flags in kebab-case (--scene-path res://main.tscn) or one --input '<json>', and summer ops describe <op> prints the exact input. The summer-cli skill has the rest.

In this skillCommand
summer_add_nodesummer engine add-node
summer_create_scenesummer engine create-scene
summer_get_scene_treesummer engine get-scene-tree
summer_import_from_urlsummer engine import-from-url
summer_instantiate_scenesummer engine instantiate-scene
summer_open_scenesummer engine open-scene
summer_save_scenesummer engine save-scene
summer_set_propsummer engine set-prop

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Scene Composition AI skill do?

Scene structure: node hierarchy, when to extract sub-scenes, reusable prefabs, instance vs add node.

Why use Scene Composition on TypingMind?

Because you install it once and use it with any model. Scene Composition is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Scene Composition in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/SummerEngine/summer/tree/main/library/skills/scene-composition. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Scene Composition?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Scene Composition?

As many as you like. As long as a model supports skills, you can use Scene Composition with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Scene Composition AI skill free?

Yes. It is published on GitHub by SummerEngine under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇