Repository navigation
Try your Project
The author confirmed that this Snake began as a university project. The experiment started with a simple prompt to explore what a modern frontier AI could do with it. This reusable prompt captures the challenge; it is not presented as a verbatim transcript of the original request.
Replace the brackets, provide the repository, and try:
This is a project I made at university: [project and purpose].
Modernize it into the best usable version you can build today, while keeping [original language] as the foundation. The original constraints were [constraints]. Preserve the idea and make the implementation something I can still study and recognize.
Start by running or inspecting the original. Explain its real limitations. Then implement the improvements: strengthen correctness, rethink the user experience, add thoughtful visual and interaction design, and go beyond a cosmetic cleanup. Use modern libraries when they help, but keep the actual application in the required language.
Make it easy to build and run. Add meaningful deterministic tests and test the interaction, failure paths and final package. Show me the actual result, not a mockup. Inspect the rendered interface and fix clipping, confusing controls and broken states. Do not claim tests or platform support you did not verify.
Write the interface and documentation in English. Explain that this is an experiment with my college project and describe the important decisions so another student can learn from them. Preserve the original through Git.
Work autonomously on reversible local changes. Ask before external publication, spending money, installing shared tools or destructive actions that I have not authorized. Finish with the runnable result, test evidence, known limitations and a concise explanation of what improved.
Give different tools the same starting commit and constraints. Record the model/tool version, date, time and follow-up prompts. A useful comparison asks:
- Does it build from a clean checkout using the documented commands?
- Does the core behavior work, including edge cases and failure paths?
- Is the experience coherent and pleasant beyond the first screenshot?
- Did it keep the language constraint and preserve an understandable design?
- Are the tests meaningful, and are limitations stated honestly?
- Can you explain the implementation after reading it?
Avoid calling one impressive demo the universally best AI. Different tools, budgets and follow-up instructions can produce different outcomes. Keep your own judgment in the loop.