Cloudkeep Rivals

Founded with GPT-5.5, made release-ready by one doctor with Codex and GPT-5.6.

Comment

Inspiration

I am a medical doctor, not a trained software engineer. I wanted to prove that a single person can keep practicing medicine while also building a real, playable game when the development process becomes a conversation with a capable engineering partner. Cloudkeep Rivals began as that experiment, with GPT-5.5 helping me establish the foundations. Build Week gave me GPT-5.6 in Codex and a focused reason to tackle the hardest part next: making the systems behind the game trustworthy enough for real players and ready for release.

What it does

Cloudkeep Rivals is a tactical fantasy strategy game set on floating islands. Players collect and upgrade heroes, develop their home island, choose battle modes, build a team, and fight lane-based encounters with distinct skills and roles.

The Build Week extension protects the game behind those visible systems. Authenticated progression now uses account-scoped server projections and receipts for wallets, energy, idle resources, buildings, heroes, battle sessions, results, claims, and seasonal progression. Guest mode remains available so a new player or judge can experience the game immediately without creating an account.

How I built it

The client is built in Unity and C#. The backend uses a Supabase Edge Function in Deno/TypeScript with PostgreSQL authority migrations and SQL regression tests.

I used GPT-5.6 inside Codex as my primary Build Week engineering collaborator. Codex mapped the existing Unity and backend code, traced trust boundaries across thousands of lines, designed the migration in testable slices, generated adversarial contract tests, implemented and reviewed client/server changes, and helped diagnose race conditions and replay risks. It also helped prepare the English demo narrative and this submission.

GPT-5.6 is not a runtime chatbot inside the game. Its role is visible in the engineering outcome: it enabled a solo non-programmer to reason about and harden a production-scale client/backend system without hiding the complexity behind a toy demo.

What is new for Build Week

  • Account/session-scoped Unity authority client
  • Server-owned wallet, energy, resources, buildings, hero training, battles, and trusted progression
  • Atomic battle start/result/claim flow instead of detached client energy spending
  • Idempotent receipts, replay protection, concurrency guards, and fail-closed validation
  • Protected cloud-save overlay that rejects client-authored economy state
  • Purchase lifecycle and retry-safe account deletion containment
  • English backend response localization
  • 133 passing Deno cross-layer contract tests, plus 48 Unity C# test files containing 319 test cases for review

Challenges I ran into

The difficult part was not adding another visible feature. It was changing who is allowed to decide that a battle was won, a reward was earned, or a resource exists without breaking the existing player experience. A mobile network can retry, time out, switch accounts, or deliver responses out of order. The migration therefore had to make every protected mutation idempotent and bind every response to the correct account, request, session, and projection version.

Another challenge was preserving guest play while making authenticated play stricter. The solution was to keep a clearly separated local guest path and require verified server authority for protected account-backed state.

Accomplishments I am proud of

I am proud that this is not a mock-up or a weekend landing page. It is a real game that can be downloaded and played. I am also proud that the Build Week work is measurable: the private review repository includes the dated source snapshot and a 133-test cross-layer suite that passes on the submitted code.

Most of all, I built this while continuing my medical work. Without ChatGPT and Codex, I would not have been able to create the game, its website, the backend, the release pipeline, and this security migration by myself.

What I learned

AI is most useful when it does more than generate isolated snippets. The largest gain came from maintaining context across product intent, client behavior, backend invariants, tests, and release constraints. I also learned that tests written as trust-boundary contracts are a powerful way for a solo builder to keep an AI-assisted codebase coherent.

What's next

Next I want to use the same account-scoped architecture to expand asynchronous competition, cooperative events, and live progression without giving the client authority over competitive outcomes. I also plan to improve onboarding analytics so I can understand where new players struggle and make the first session clearer without making the strategy shallower.

Built With

Share this project:

Updates