How to use Claude to build a startup ranking game
Claude Code and Codex built a startup-ranking game. Review moved ELO to the server and fixed reset timing while the owner chose mechanics and launched a beta.
rossgarlick.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Build this startup-comparison game with server-side ELO and explicit authorization around score changes. Review the daily reset's UTC boundary and retry/idempotence behavior, keep credentials out of chat and client code and give me reviewable tests before the beta launch.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Ross Garlick
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Garlick launched a public beta with server-side ELO and corrected reset behavior. Product mechanics and operational decisions remained under human review.
- Human involvement
- Garlick chose product mechanics, checked the interface and reviewed operational corrections including credentials and reset behavior.
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
- Garlick converted research and Cowork planning files into a Claude Code implementation handoff.
- Claude built the game while Codex reviewed scoring and operational robustness.
- They moved ELO to the server and corrected reset timing before the public beta.
Ross Garlick used Claude Code and Codex to build Coliseo, a live game for comparing and ranking startups. Planning began with research and local documents in Cowork, then moved to Claude Code when a GitHub connector stalled the handoff.
Claude Code built the application around GitHub, Supabase and Vercel. Garlick continuously chose mechanics and checked the interface, while Codex acted as another reviewer for robustness. The game evolved beyond a simple pairwise ranking screen into a king-of-the-hill mode and matchmaking intended to distinguish conviction from mere familiarity with a famous company.
Several corrections affected whether the game could be trusted. ELO calculations moved to the server so players could not alter scores on the client. A bulk reset query replaced a less suitable reset approach, and the team corrected the UTC midnight reset. Garlick also rotated a database password after it appeared in the conversation.
The source reports a public beta and a configured domain, not only a generated prototype. It also documents the human product decisions and operational review behind that launch. The useful pattern is an implemented ranking game with a second agent reviewing its scoring and reset boundaries, while the owner remained responsible for deciding how the competition should work.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Personal blog · Ross Garlick
Frequently asked questions
Why move ELO calculations to the server?
Client-side calculations allowed players to tamper with scores.
What reset behavior was corrected?
The bulk reset approach and its UTC midnight timing.
What role did Codex play?
It reviewed robustness alongside Claude Code's implementation work.
More details
About this account
Garlick converted research and Cowork planning files into a Claude Code implementation handoff. Claude built the game while Codex reviewed scoring and operational robustness.
Original beta-launch account opened. Development involved Claude Code and Codex, with Cowork planning. Reported speed is not an independently measured comparison. Suggested prompt written by Agents In Use; no original task prompt was published in the verified account.
Credit: Ross Garlick — Personal blog
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · Codex · Supabase · Vercel · GitHub
Ranking game · Server-side scoring
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.