unity-rapid-prototyping

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

When to Use

使用场景

Planning or implementing any feature, mechanic, or system in a Unity game (jam or prototype)
An agent is about to write C# or build scenes via the Unity MCP
Deciding between building something vs. using an existing asset, package, or placeholder
Reviewing scope when time is limited and the game must ship

在Unity游戏（game jam或原型）中规划或实现任何功能、机制或系统时
即将通过Unity MCP编写C#代码或构建场景时
纠结是自行开发还是使用现有资源、包或占位符时
时间有限且必须完成游戏时的范围评审

Prerequisites — validate the environment BEFORE anything

前置条件——先验证环境

The skill only works when the agent can reach Unity through the MCP bridge. Run this checklist once per session and report the result:

Check	How to validate	Required
Unity Editor connected via MCP	`unityMCP` tools respond; editor state ready	✅
`com.coplaydev.unity-mcp` installed	`manage_packages(action="list_packages")` shows it (bridge from https://github.com/CoplayDev/unity-mcp)	✅
`com.unity.ai.assistant` installed	`manage_packages(action="list_packages")` shows it (Unity AI Agents)	✅
`com.unity.test-framework` installed	`manage_packages(action="list_packages")` shows it	✅ for EditMode tests

If any required package is missing, STOP and tell the user how to install it — do NOT start building. Do not guess package names: read the actual package list.

只有当Agent能通过MCP桥接访问Unity时，本技能才能生效。每次会话运行一次以下检查清单并报告结果：

检查项	验证方式	是否必需
Unity Editor通过MCP连接	`unityMCP` 工具响应；编辑器状态就绪	✅
`com.coplaydev.unity-mcp` 已安装	`manage_packages(action="list_packages")` 显示该包（桥接地址：https://github.com/CoplayDev/unity-mcp）	✅
`com.unity.ai.assistant` 已安装	`manage_packages(action="list_packages")` 显示该包（Unity AI Agents）	✅
`com.unity.test-framework` 已安装	`manage_packages(action="list_packages")` 显示该包	✅（仅EditMode测试需要）

如果任何必需包缺失，请停止操作并告知用户安装方法——不要开始构建。不要猜测包名称：请查看实际包列表。

Install commands (when missing)

安装命令（当包缺失时）

bash

undefined

bash

undefined

unity-mcp (CoplayDev) — Package Manager → Add from git URL

unity-mcp (CoplayDev) — 包管理器 → 从Git URL添加

manage_packages(action="add_package", package="https://github.com/CoplayDev/unity-mcp.git?path=/MCPForUnity#main")

Unity AI Assistant (Agentes.IA) — requires Unity 6.0+, project linked to Unity Cloud

Unity AI Assistant (Agentes.IA) — 需要Unity 6.0+，项目已关联Unity Cloud

manage_packages(action="add_package", package="com.unity.ai.assistant")

Test Framework (for EditMode/PlayMode validation)

Test Framework（用于EditMode/PlayMode验证）

manage_packages(action="add_package", package="com.unity.test-framework")

undefined

manage_packages(action="add_package", package="com.unity.test-framework")

undefined

Idea Validator — from theme to ONE playable concept

想法验证器——从主题到可玩概念

The jam starts with a theme, not a game. Go from theme → ONE direction whose core loop is buildable in the jam time. This is a KILLER of ideas, not a cheerleader: it says NO to ~90% of ideas.

Game jam从主题开始，而非游戏。从主题出发，确定一个能在jam时间内完成核心循环的方向。这是想法的“筛选器”，而非“啦啦队”：会否决约90%的想法。

Collect preferences first — ask, don't assume

先收集偏好——询问而非假设

The theme shapes the idea, but the human shapes the choices. ASK the user's preferences BEFORE generating directions — do NOT invent their skills. Keep it to 5 questions max, ask one at a time:

#	Question	Why it matters
1	What genre are you strongest at / enjoy most? (e.g. shooter, platformer, puzzle)	Strongest genre = highest odds of finishing well
2	How much jam time do you have? (hours)	Sets the scope ceiling for "Finish in jam time"
3	Solo or team? Team size?	More hands = more scope allowed
4	Unity + C# comfort level? (beginner / comfortable / strong)	Beginner → simpler architecture + more placeholders
5	Any must-haves or must-avoids? (e.g. no 3D, love roguelikes, want local multiplayer)	Hard constraints that can kill an idea instantly

Rules:

Preferences are a FILTER, not the whole game: genre narrows the field, theme shapes the idea. A shooter preference + theme "growth" → a shooter about growing, not just a shooter.
If the user does not answer, use sensible defaults (top-down action, 48h, solo, comfortable) and STATE the assumption.
When generating directions, honor hard preferences (genre, must-avoids) in EVERY pitch; use soft ones (time, team, skill) to tune scope.

主题塑造想法，但用户的选择决定方向。在生成方向前先询问用户偏好——不要臆测他们的技能。最多问5个问题，一次问一个：

序号	问题	重要性
1	你最擅长/最喜欢的游戏类型是什么？（例如：射击、平台、解谜）	擅长的类型=更高的完成质量概率
2	你有多少jam时间？（小时）	设定“在jam时间内完成”的范围上限
3	单人还是团队？团队规模？	人手越多=允许的范围越大
4	Unity + C#的熟练程度？（新手/熟练/精通）	新手→更简单的架构+更多占位符
5	有没有必须包含或必须避免的内容？（例如：不做3D、喜欢肉鸽、想要本地多人）	硬性约束可直接否决想法

规则：

偏好是筛选器，而非全部：类型缩小范围，主题塑造想法。比如偏好射击+主题“成长”→一款关于成长的射击游戏，而非单纯的射击游戏。
如果用户未回答，使用合理默认值（俯视角动作、48小时、单人、熟练）并说明假设。
生成方向时，每个提案都要遵守硬性偏好（类型、必须避免的内容）；使用软性偏好（时间、团队、技能）调整范围。

Generate 3–5 directions

生成3-5个方向

Give every direction a one-sentence pitch: genre + the exact 10-second action the player does + how it interprets the theme.

Direction	One-sentence pitch	Theme take (literal / metaphor)
A	...	...
B	...	...
C	...	...

每个方向用一句话概括：类型+玩家10秒内的具体动作+主题诠释方式。

方向	一句话概括	主题解读（字面/隐喻）
A	...	...
B	...	...
C	...	...

Validate each direction

验证每个方向

Criterion	Question	Fail →
Core loop	Can you state what the player does every 10 seconds in ONE sentence?	Not ready — rephrase or cut.
Prototype in <1h	Can the core loop be grey-boxed in 1 hour?	Too many systems — simplify or cut.
Fun-risk	Is the fun testable in play mode immediately (1 object, 1 target)?	Defer — needs 5 systems to feel anything.
Finish in jam time	Can it be FINISHED (not just started) in the jam hours?	Cut or shrink scope.
Standout	What makes it memorable vs 100 entries with the same theme?	No answer — it's the crowd idea, rethink.
Theme tie	Can you defend the theme connection in one sentence?	Literal ties are safe but crowded; metaphor is riskier but more memorable.

标准	问题	不通过→
核心循环	你能用一句话说明玩家每10秒做什么吗？	未准备好——重新表述或删减。
1小时内原型化	核心循环能在1小时内完成灰盒制作吗？	系统过多——简化或删减。
趣味风险	玩法趣味能立即在PlayMode中测试吗？（1个对象、1个目标）	延后——需要5个系统才能感受到趣味。
在jam时间内完成	能在jam时间内完成（而非仅启动）吗？	删减或缩小范围。
独特性	与同主题的100个作品相比，它的记忆点是什么？	无答案——属于大众想法，重新构思。
主题关联	你能用一句话捍卫与主题的关联吗？	字面关联安全但同质化；隐喻关联风险更高但更具记忆点。

Verdicts

结论

BUILD NOW — passes everything: start the mini-prototype in the first hour.
VALIDATE FIRST — fun is uncertain: build only the core loop grey-box; fun in that hour decides if it stays.
CUT — fails core loop, scope, or standout: record it, move to the next direction.

立即构建——通过所有验证：在第一个小时开始制作迷你原型。
先验证——趣味不确定：仅构建核心循环灰盒；一小时内的趣味测试决定是否保留。
否决——未通过核心循环、范围或独特性测试：记录下来，转向下一个方向。

Rules

规则

The first hour of the jam is a MINI-PROTOTYPE of the chosen idea. If it is not fun after that hour, switch to the #2 direction. Do NOT polish it.
Do not fall in love with an idea. The validator says NO until an idea survives every criterion.
Division of labor: the AI proposes directions and enforces constraints (quantity + discipline); the human owns taste and the final pick. The AI is the strict mentor, NOT the creative brain.

Jam的第一个小时是所选想法的迷你原型。如果一小时后仍无趣味，切换到第2个方向。不要打磨。
不要沉迷于某个想法。验证器会否决所有想法，直到有一个通过所有标准。
分工：AI提出方向并执行约束（数量+纪律）；用户负责审美和最终选择。AI是严格的导师，而非创意大脑。

Critical Patterns

关键模式

1. The Decision Gate — run it BEFORE proposing any code

1. 决策闸门——在提议任何代码前运行

For EVERY feature request, answer these five questions internally, in order, and show the verdict when you propose:

#	Question	Pass	Fail → do this
1	Does it serve the CORE LOOP?	It directly supports the primary gameplay loop	Don't build now. Write it down, move on.
2	Can it be done in <1 hour?	Small, single-mechanic scope	Split it, cut it, or timebox it.
3	Is there a 10x simpler version?	The simplest possible version is identified	Keep simplifying until the 10x version is obvious.
4	Asset over custom?	An existing asset/placeholder is good enough	Use the asset. Do NOT hand-roll.
5	Does it raise the odds of FINISHING?	The game gets closer to playable	Cut it. Finishing beats polish.

A feature that fails the gate is not wrong — it is NOT NOW. Record it and protect the core loop.

对于每个功能请求，按顺序内部回答以下5个问题，并在提议时给出结论：

序号	问题	通过标准	不通过→执行操作
1	它服务于核心循环吗？	直接支持主要游戏循环	现在不构建。记录下来，继续下一步。
2	能在1小时内完成吗？	范围小，单一机制	拆分、删减或设定时间盒。
3	有没有10倍简化的版本？	已确定最简单的版本	持续简化直到10倍版本明确。
4	优先使用资源而非自定义？	现有资源/占位符足够用	使用资源。不要自行开发。
5	它能提高完成概率吗？	游戏更接近可玩状态	删减。完成度优于打磨。
未通过闸门的功能并非错误——只是“现在不做”。记录下来并保护核心循环。

2. Gameplay-first mandate

2. 玩法优先要求

Validate the fun BEFORE the architecture. A grey-boxed mechanic in play mode beats a clean-but-unplayable system.
Minimal architecture: no DI containers, no event buses, no "extensible systems" in a jam. One scene, plain MonoBehaviours, direct references.
Placeholders are features: cubes, capsules, and white sprites are acceptable stand-ins for art until the loop is fun.

在架构前验证趣味。PlayMode中的灰盒机制优于干净但不可玩的系统。
极简架构：jam中不使用DI容器、事件总线或“可扩展系统”。单个场景、普通MonoBehaviours、直接引用。
占位符就是功能：立方体、胶囊体和白色精灵在循环有趣前可作为美术的替代。

3. Kill-or-simplify timebox

3. 否决或简化时间盒

If a mechanic is not fun after one focused iteration, kill it or simplify it. Do NOT polish a bad idea.
Prefer deleting code over refactoring it when the feature is not core.

如果一个机制经过一次专注迭代后仍无趣味，否决或简化它。不要打磨糟糕的想法。
当功能非核心时，优先删除代码而非重构。

Unity MCP Workflow

Unity MCP工作流

Idea validated first: ask the user's preferences (genre, time, team, skill, must-haves) → run the Idea Validator with theme + preferences → ONE direction with a core loop stated in one sentence. Do NOT open the editor before this exists.
Validate prerequisites first (see Prerequisites section): editor connected via MCP + required packages installed. Then check
```
read_console
```
for compile errors — never build on a broken compile.
Create or modify scripts, then
```
read_console
```
immediately to catch compilation errors before proceeding.
Validate visual/gameplay changes with play mode (
```
manage_editor
```
play/pause/stop) + screenshots; validate logic with
```
run_tests
```
(EditMode/PlayMode).
Build the scene with the minimum GameObjects needed: camera, light, one playable object, one interactable target.
Iterate: change → compile check → play → confirm fun → commit.

先验证想法：询问用户偏好（类型、时间、团队、技能、必须包含的内容）→结合主题和偏好运行想法验证器→确定一个能用一句话说明核心循环的方向。在此之前不要打开编辑器。
先验证前置条件（见前置条件部分）：编辑器通过MCP连接+必需包已安装。然后检查
```
read_console
```
是否有编译错误——永远不要在编译错误的基础上构建。
创建或修改脚本，然后立即
```
read_console
```
以在继续前捕获编译错误。
通过PlayMode（
```
manage_editor
```
的播放/暂停/停止）+截图验证视觉/玩法变化；通过
```
run_tests
```
（EditMode/PlayMode）验证逻辑。
使用最少的GameObject构建场景：相机、灯光、一个可玩对象、一个可交互目标。
迭代：修改→编译检查→播放→确认趣味→提交。

Code Examples

代码示例

Minimal before polished — the whole game loop in a handful of MonoBehaviours:

csharp

// Placeholder player controller — enough to validate the core loop
public class Player : MonoBehaviour
{
    public float speed = 5f;

    void Update()
    {
        var h = Input.GetAxis("Horizontal");
        var v = Input.GetAxis("Vertical");
        transform.position += new Vector3(h, 0, v) * speed * Time.deltaTime;
    }
}

Pure logic, easily testable in EditMode — keep game rules outside MonoBehaviours so they can be validated fast:

csharp

public class Health
{
    public int Current { get; private set; }

    public Health(int start) => Current = start;

    public void TakeDamage(int amount) => Current = Mathf.Max(0, Current - amount);
}

csharp

// Assets/Tests/EditMode/HealthTests.cs
public class HealthTests
{
    [Test]
    public void Health_NeverGoesBelowZero()
    {
        var h = new Health(5);
        h.TakeDamage(99);
        Assert.AreEqual(0, h.Current);
    }
}

极简优先于精致——少量MonoBehaviours实现完整游戏循环：

csharp

// 占位符玩家控制器——足以验证核心循环
public class Player : MonoBehaviour
{
    public float speed = 5f;

    void Update()
    {
        var h = Input.GetAxis("Horizontal");
        var v = Input.GetAxis("Vertical");
        transform.position += new Vector3(h, 0, v) * speed * Time.deltaTime;
    }
}

纯逻辑，易于在EditMode中测试——将游戏规则放在MonoBehaviours之外以便快速验证：

csharp

public class Health
{
    public int Current { get; private set; }

    public Health(int start) => Current = start;

    public void TakeDamage(int amount) => Current = Mathf.Max(0, Current - amount);
}

csharp

// Assets/Tests/EditMode/HealthTests.cs
public class HealthTests
{
    [Test]
    public void Health_NeverGoesBelowZero()
    {
        var h = new Health(5);
        h.TakeDamage(99);
        Assert.AreEqual(0, h.Current);
    }
}

Commands

命令

bash

undefined

bash

undefined

Unity MCP: validate environment first

Unity MCP：先验证环境

manage_packages(action="list_packages") # required: com.coplaydev.unity-mcp, com.unity.ai.assistant, com.unity.test-framework

manage_packages(action="list_packages") # 必需：com.coplaydev.unity-mcp, com.unity.ai.assistant, com.unity.test-framework

Unity MCP: validate logic fast

Unity MCP：快速验证逻辑

run_tests(mode="EditMode") # then poll get_test_job

run_tests(mode="EditMode") # 然后轮询get_test_job

Unity MCP: validate feel

Unity MCP：验证手感

manage_editor(action="play") # play mode read_console(types=["error"]) # compile + runtime errors

undefined

manage_editor(action="play") # 播放模式 read_console(types=["error"]) # 编译+运行时错误

undefined

Resources

资源

Unity MCP tools: full tool list in the session (unityMCP server).
Decision Gate: keep the 5-question table in front of every proposal.

Unity MCP工具：会话中的完整工具列表（unityMCP服务器）。
决策闸门：在每个提案前参考5问题表格。