How to use Codex to fix a puzzle game’s reward ledger

Fatuma Yattani opened the completed Faultline game in Codex for a final engineering review. Players inspect two answers, locate the flawed sentence, diagnose the mistake, and choose a verification method. That staged mechanic made scoring and reward history more involved than a simple quiz.

Codex found that a fast player could receive a clean-round bonus despite an earlier mistake, and replays could award extra experience. Daily reward enforcement also depended on bounded result history: enough later rounds could remove evidence that a reward had already been collected.

The repair separated the dated reward ledger from general history. Tests covered same-day replay, history eviction, next-day rewards, and malformed older progress. Other changes handled damaged browser storage and restored keyboard focus after interface rerenders. Codex expanded the reported passing suite from twelve cases to eighteen.

Yattani reviewed the implementation plan, approved its scope, reran checks independently, and requested another review before merging. That additional review caught another reward-history edge case. The result was a more robust playable game with explicitly checked reward rules. It was not a runtime AI feature: Codex’s role was finding and repairing implementation defects around a human-designed game.

Prompt
Suggested prompt (written by Agents In Use, not the author's)

Source: Devpost · Fatuma Yattani