Godot Brainstorming

A structured design process for Godot 4.3+ features and systems — from blank slate to a clear scene tree, signal map, and data flow before you write a single line of implementation code.

Related skills: godot-grill for settling open design decisions first, scene-organization for scene tree composition patterns, component-system for component-based architecture, event-bus for signal-based communication design.

---

Process: How to Brainstorm

Do NOT jump straight to designing. Follow these steps:

Step 1: Settle the decisions

If the request has open design decisions (scope, dimension, authority, data home, …), invoke godot-prompter:godot-grill and let it run to its end. Skip it when a record in the project's decisions or ADR directory already covers this feature, or the user has stated the decisions. Either way, check what already exists (code, scenes, assets). Carry the record into Step 2 — approaches must respect its settled rows.

Step 2: Propose 2-3 approaches

With the decisions settled, propose architectural options with trade-offs. For example:

• "Enum FSM vs Node FSM for your state machine — here's when each fits"
• "EventBus vs direct signals for your systems — here's the trade-off" Lead with your recommendation and explain why.

Step 3: Design with approval

Present the design section by section (scene tree, signal map, data flow). Ask "does this look right?" after each section before continuing.

Step 4: Prepare for implementation

After the design is approved:

1. Offer the agent instructions section — Ask whether to add the GodotPrompter integration section to the file this project already uses for agent instructions (see Agent Instructions Injection below). It is what tells subagents and future sessions to use GodotPrompter skills. Skip silently if a ## GodotPrompter section already exists in any of them, or if the user has declined before. Never add it without agreement.

2. Create implementation plan — If a planning skill is available (e.g., superpowers:writing-plans), use it; if not, break the design into ordered tasks yourself. Either way the project decides where the plan is saved: the user's instructions or the project's agent instructions file; then the planning skill's convention; then an existing plans directory. If none applies, ask the user, suggesting docs/plans/, or use docs/plans/ when you cannot ask.

3. Annotate each task with skills — Every task in the plan that involves a Godot system MUST list which godot-prompter:* skill(s) to invoke during implementation. Example:

   • Task 3: Player movement — Create CharacterBody3D with walk, sprint, jump. Skills: godot-prompter:player-controller, godot-prompter:input-handling

   This ensures that even when another plugin executes the plan, the implementing agent knows which GodotPrompter skills to load.

---

1. When to Use

Start here whenever you are:

• Adding a new feature — a chest, a dialogue system, a crafting bench, a skill tree
• Creating a new scene — you need to decide what nodes it contains and how they communicate
• Choosing between approaches — inheritance vs. composition, Autoload vs. Resource, 2D vs. 3D
• Feeling stuck on structure — the code works but the scene tree feels wrong
• Onboarding someone — you need to explain the design of an existing system

If you already know exactly what nodes you need and how they connect, skip this skill and build. Use it when uncertainty is slowing you down.

---

2. Scene Tree Planning

Sketch the scene tree on paper (or in a comment block) before opening the Godot editor. The goal is to answer three questions for every node:

1. What does this node own? (data, child nodes, visual representation)
2. What does this node do? (its single responsibility)
3. How does it talk to neighbors? (signals up, method calls down, EventBus sideways)

Planning Steps

1. Name the root node and its type — this defines the scene's contract with the world.
2. List immediate children by responsibility group, not by Godot node type.
3. Assign a Godot node type to each entry.
4. Identify every signal the scene emits and every signal it consumes.
5. Mark which nodes should be separate .tscn files (reuse candidates).

Example: Planning a "Chest" Interactable

Step 1 — Name and root type

A Chest is a world object the player walks up to and opens. It is not a physics body; it does not move. Root: StaticBody2D or Node2D.

Step 2 — Responsibility groups

• Visual representation (sprite, animation)
• Collision / interaction trigger (detect player proximity)
• Loot data (what items are inside)
• UI feedback (prompt label, open animation trigger)
• State (is it open or closed?)

Step 3 — Assign node types

Chest (StaticBody2D)
├── Sprite2D                  # closed/open frame, or swap texture on open
├── AnimationPlayer           # open animation
├── CollisionShape2D          # physical body shape (blocks player)
├── InteractionArea (Area2D)  # detect when player is close enough
│   └── CollisionShape2D      # slightly larger than body shape
├── PromptLabel (Label3D or Label) # "Press F to open"
└── LootTable (Node)          # holds @export var items: Array[ItemData]

Step 4 — Signal map

Signal	Emitted by	Connected to	Purpose
body_entered(body)	InteractionArea	Chest._on_area_body_entered	Show prompt when player enters range
body_exited(body)	InteractionArea	Chest._on_area_body_exited	Hide prompt when player leaves
opened(loot: Array[ItemData])	Chest	InventorySystem or EventBus	Deliver loot to whoever owns the inventory
animation_finished(name)	AnimationPlayer	Chest._on_animation_finished	Lock chest after open animation completes

Step 5 — Reuse candidates

LootTable is likely reused by barrels, enemies, and shop crates — extract it as a separate .tscn component.

For the resulting GDScript and C# Chest sketches, plus the four-part design entry (Scene Tree, Node Responsibilities, Signal Map, Data Flow) used to document this design, see references/example-chest.md.

---

3. Picking Node Types and Dimension

Two lookups belong here but are pure recall — load them only when the answer is not already obvious:

• Which node for which need? CharacterBody vs RigidBody vs StaticBody vs Area, UI vs world-space labels, particles, cameras, spawn markers.
• 2D, 3D, or 2.5D? Selection criteria for each, hybrid techniques (billboarded sprites, orthographic 3D, SubViewport UI), and the performance consequences.

Two Godot 4.3+ specifics are easy to get wrong and worth stating up front: tile-based levels use TileMapLayer (one layer per node — TileMap is deprecated), and blend-tree locomotion needs an AnimationTree paired with an AnimationPlayer, not an AnimationPlayer alone.

Full need-to-node table, the 2D/3D decision criteria, and 2.5D hybrid techniques: references/node-selection.md

---

4. Questions to Ask Before Building

Work through this checklist before creating your first node.

• What data does this system need? — List every piece of state: position, health, item count, flags
• Who owns each piece of data? — Assign one authoritative owner per value; avoid duplicating state
• How does it communicate? — Signals up the tree, method calls down, EventBus for cross-system events
• Can it be reused? — If yes, it should be a separate .tscn scene with a clean @export interface
• Does it need persistence? — If the data must survive scene changes or game restarts, plan a save system early
• What is the scene tree? — Sketch at least two levels deep before touching the editor
• What signals does it emit? — List every signal name, its arguments, and who connects to it
• What are the failure modes? — What happens if a required node is missing? If a signal fires twice?
• What is the minimum viable version? — Build that first; add complexity only when it is needed

---

5. Common Architecture Decisions

If you need...	Consider...	Why
Global state accessible anywhere	Autoload (singleton)	Registered in Project Settings; available as a named global
Data shared between multiple scenes	Resource (.tres / .res)	Saved as an asset; @export-able; survives scene reloads
Reusable behavior across entity types	Component scene	Instantiate as a child; each entity opts in by including the scene
Complex entity behavior with many states	State machine	Explicit enter/exit per state; prevents if-chain sprawl
Events between systems that don't share a parent	EventBus Autoload	Decouples sender and receiver; any node can connect
Data that must persist across sessions	Save system with JSON or binary	Serialize Resource or Dictionary; load on _ready
Configurable game data (stats, items, levels)	Resource with @export fields	Edit values in the Inspector; no code change required
Spawning scenes at runtime	PackedScene + instantiate()	Store @export var scene: PackedScene; call scene.instantiate()
Running code on a delay or interval	Timer node	Cleaner than _process frame counters; supports one-shot and loop
Gradual transitions (fade, lerp, tween)	Tween	create_tween() is built-in; no extra node required in Godot 4

---

6. Design Output Format

Capture your design in a comment block at the top of the root script, or in a DESIGN.md file next to the scene. A complete design entry has four parts: a scene tree ASCII diagram, a node responsibilities table, a signal map (signal → source → consumer → payload), and a data flow trace showing how a triggering event propagates through the tree.

See references/example-chest.md for a fully worked four-part entry built around the Chest interactable.

---

Design Checklist

• Scene tree sketched at least two levels deep before opening the editor
• Every node has a single named responsibility
• All signals listed with name, source, consumer, and payload type
• Data ownership assigned — no value stored in two places
• Reuse candidates extracted to separate .tscn files
• Communication pattern chosen: signals up, calls down, EventBus sideways
• Persistence requirements identified before building data structures
• Architecture decision table consulted for global state, shared data, and events

---

Agent Instructions Injection

When preparing for implementation (Step 4), add the following section to the file this project uses for agent instructions. Check for an existing ## GodotPrompter heading in all of CLAUDE.md, CLAUDE.local.md, .claude/CLAUDE.md, AGENTS.md, GEMINI.md, .github/copilot-instructions.md, and the .claude/rules/ and .cursor/rules/ directories — if it is present anywhere, skip.

A previous refusal counts as an answer. The SessionStart hook records one as "section_offer": "declined" in this project's ~/.godot-prompter/state/<hash>.json (godot-prompter:godot-mentor documents the file and how the hash is computed) — if that key is set, do not ask again. Record a refusal there yourself, merged into any existing keys, when the user declines here.

Write it to the file the project already maintains, not to a second one:

• CLAUDE.md exists → add it there.
• Only AGENTS.md or GEMINI.md exists → add it there, so the repo stays agent-agnostic. Claude Code reads CLAUDE.md only, so if the user wants it loaded there too, a one-line CLAUDE.md containing @AGENTS.md does that — offer it, do not assume it.
• Nothing exists → create CLAUDE.md with a project overview header and this section.

GodotPrompter

This is a Godot project with GodotPrompter skills available. Before implementing any game system, you MUST check for a matching godot-prompter:* skill and invoke it. This applies to all agents, subagents, and sessions working in this repository.

Key skills: player-controller, state-machine, event-bus, scene-organization, component-system, resource-pattern, godot-ui, hud-system, ai-navigation, camera-system, audio-system, save-load, inventory-system, godot-testing.

For the full skill list, invoke godot-prompter:using-godot-prompter.