How to use Codex to verify a deterministic tower RPG
Codex extended a deterministic tower RPG with resource-aware solvability checks and complete-state rewind; its creator retained balance and release review.
devpost.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Extend this deterministic tower RPG while sharing authoritative rules among gameplay, forecasts, saves, and the solver. Model keys, shops, equipment, build choices, rank, and backtracking. Make rewind restore complete decision state and consume one charge safely across autosaves. Test web/desktop seed parity and viable routes, then present balance and playtest gaps for my review.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Raymond Chan
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The project reached a release-tested ten-floor web and desktop game with checked deterministic replay, complete-state rewind, and solver-verified strategic routes.
- Human involvement
- Raymond defined rules and balance, approved visuals and acceptance criteria, reviewed every assisted change, and controlled release decisions.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Codex extended the three-floor foundation with strategic progression, build choices, and rewind mechanics.
- It aligned combat, reducer state, saves, and a resource-aware solver using shared authoritative data.
- Chan reviewed design and release evidence while retaining balance and visual acceptance decisions.
Raymond Chan brought a playable three-floor prototype into OpenAI Build Week and used Codex to extend it into Crown Above the Veil, a ten-floor tactical puzzle RPG. The new systems included distinct build choices, deliberate backtracking, scarce resources, and a limited rewind mechanic.
Because combat was deterministic, the displayed forecast and production rules had to agree exactly. Codex worked on reducer transitions, combat traits, progression, save state, regression cases, and a resource-aware solver. That solver needed to model shops, keys, equipment, doors, rank milestones, and floor transitions rather than approving an imaginary route.
Rewind had a similar constraint: restoring position alone was insufficient. The game needed to restore the complete decision state while consuming a charge and staying consistent with autosaves. Shared authoritative data kept web, desktop, and solver behavior aligned.
Chan retained rules, balance targets, visual approval, acceptance criteria, and final review. The account reports a release-tested web and desktop game with viable routes for its build choices and deterministic replay checks. Broader blind playtesting remained future work. Automated evidence showed that a path existed; it did not answer whether every new player would find the clues fair or the backtracking memorable.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Devpost · Raymond Chan
Frequently asked questions
What did the solver need to model?
Resources, keys, doors, shops, equipment, rank, builds, and floor transitions.
Why is position-only rewind insufficient?
A decision also changes resources, progression, and saved state.
Can solvability checks prove the game feels fair?
No. The author explicitly planned broader human blind playtesting.
More details
About this account
Codex extended the three-floor foundation with strategic progression, build choices, and rewind mechanics. It aligned combat, reducer state, saves, and a resource-aware solver using shared authoritative data.
Original solo-developer account; early Claude prototype is separated from Codex extension. Reported solver and release checks are not independently rerun. Blind player comprehension remains a stated future validation step. Suggested prompt written by Agents In Use; no original task prompt was published in the verified account.
Credit: Raymond Chan — Devpost
Source published
Added to Agents In Use
Source checked
Tools and topics
Codex · TypeScript · PixiJS · Electron · Vitest · solver
deterministic games · solvability · rewind state
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.