Prototype

是什么

帮你在拍板做一个新概念之前，用最低成本造一个能跑的原型来回答"这个想法到底成不成立"。它不追求工程质量，追求的是"一次性回答一个具体问题"，并出一份结构化的原型小结。

怎么用

1. 用一句话写清楚这次原型要回答的核心问题，不要写成需求清单。
2. 让它扫一眼当前的技术栈与已有素材，确认原型用什么底子搭最快。
3. 给它定一个明确的时间盒（比如半天、一天），别让原型滑成小项目。
4. 跑完原型后，让它输出一份结构化小结：核心问题、答案、证据、下一步建议。
5. 如果结论是"成立"，再决定要不要立项；如果结论是"不成立"，把这份小结归档进决策日志。

架构图

flowchart LR
  A[一个核心问题] --> B[最小可行原型方案]
  B --> C[时间盒内搭建]
  C --> D[原型验证结果]
  D --> E[结构化决策小结]

When this skill is invoked:

1. Read the concept description from the argument. Identify the core question this prototype must answer. If the concept is vague, state the question explicitly before proceeding.

2. Read CLAUDE.md for project context and the current tech stack. Understand what engine, language, and frameworks are in use so the prototype is built with compatible tooling.

3. Create a prototype plan: Define in 3-5 bullet points what the minimum viable prototype looks like. What is the core question? What is the absolute minimum code needed to answer it? What can be skipped?

4. Create the prototype directory: prototypes/[concept-name]/ where [concept-name] is a short, kebab-case identifier derived from the concept.

5. Implement the prototype in the isolated directory. Every file must begin with:

   // PROTOTYPE - NOT FOR PRODUCTION
   // Question: [Core question being tested]
   // Date: [Current date]
   

   Standards are intentionally relaxed:

   • Hardcode values freely
   • Use placeholder assets
   • Skip error handling
   • Use the simplest approach that works
   • Copy code rather than importing from production

6. Test the concept: Run the prototype. Observe behavior. Collect any measurable data (frame times, interaction counts, feel assessments).

7. Generate the Prototype Report and save it to prototypes/[concept-name]/REPORT.md:

## Prototype Report: [Concept Name]

### Hypothesis
[What we expected to be true -- the question we set out to answer]

### Approach
[What we built, how long it took, what shortcuts we took]

### Result
[What actually happened -- specific observations, not opinions]

### Metrics
[Any measurable data collected during testing]
- Frame time: [if relevant]
- Feel assessment: [subjective but specific -- "response felt sluggish at
  200ms delay" not "felt bad"]
- Player action counts: [if relevant]
- Iteration count: [how many attempts to get it working]

### Recommendation: [PROCEED / PIVOT / KILL]

[One paragraph explaining the recommendation with evidence]

### If Proceeding
[What needs to change for a production-quality implementation]
- Architecture requirements
- Performance targets
- Scope adjustments from the original design
- Estimated production effort

### If Pivoting
[What alternative direction the results suggest]

### If Killing
[Why this concept does not work and what we should do instead]

### Lessons Learned
[Discoveries that affect other systems or future work]

8. Output a summary to the user with: the core question, the result, and the recommendation. Link to the full report at prototypes/[concept-name]/REPORT.md.

Important Constraints

• Prototype code must NEVER import from production source files
• Production code must NEVER import from prototype directories
• If the recommendation is PROCEED, the production implementation must be written from scratch -- prototype code is not refactored into production
• Total prototype effort should be timeboxed to 1-3 days equivalent of work
• If the prototype scope starts growing, stop and reassess whether the question can be simplified