Example project · A game built from scratch
Claude Cope: a game built from scratch with ProPR.
A browser game written by AI agents from the first file to the live site. The repository was empty on 1 April 2026. As of 11 November 2026 it holds 458 merged pull requests, 412 of them written by agents through ProPR.
Overview
What it is
Claude Cope is a multiplayer terminal game that runs in the browser. It parodies AI coding assistants. Players take absurd tickets from a shared backlog, argue with an AI that refuses to help, and earn "Technical Debt" for finishing. Its README says the game doubles as a stress test for ProPR. Play it at cope.bot.
Where it started
The repository was empty until the first merged pull request, #11 on 1 April 2026, set up the project's root configuration. The terminal, the ticket backlog, the leaderboard, sign-in, the ASCII-art buddies and the rate limits all arrived the same way, as pull requests you can open and read.
Most of the game took shape in one month: 399 pull requests merged in April 2026 alone.
How the work ran through ProPR
- A plan became issues.An issue is a written request for one piece of work. ProPR reads a description of what you want and splits it into issues small enough to finish in one go. The first ones here were numbered phases, starting with the project skeleton.
- Each issue became a pull request.An agent took the issue, worked on its own branch, and sent back a pull request with the proposed change laid out line by line. 412 of the 458 merged pull requests were written this way. Claude Opus 4.6 wrote 330 of them, GPT-5.4 wrote 48 and Claude Opus 4.5 wrote 30.
- Review, then merge.A person read each pull request before it landed. Related ones were grouped into epics, one branch per feature where its pull requests collect, and 45 pull requests carry "Epic" in the title. Nothing reaches the live game until a pull request is merged.
The numbers, as of 11 November 2026.
Agents that wrote the pull requests: Claude Opus 4.6 (330 pull requests), GPT-5.4 (48), Claude Opus 4.5 (30), plus a few others. 418 issues labelled AI have been closed. Counts come from the public repository on GitHub.
Four pull requests from the repository
The titles below are quoted from the public repository. Open one to see the issue it came from, the change it made and the review it went through.
- Phase 0: Initialize Monorepo Root Configuration#11
- Implement ANSI Terminal Animations for Achievements#286
- Upgrade Buddies to Multi-line ASCII Art#302
- Widen the leaderboard overlay by 50% so long usernames and vanity titles fit#870
The last one shows the everyday shape of the work: a player name did not fit, so a small pull request widened the overlay, and changes of that size make up most of any product.
What this shows for your own project
An empty repository is the simplest starting point, because there is no old code to understand and the plan is the whole product. Write down what you want, let ProPR split it into issues, and the product assembles one reviewed pull request at a time.
When a pull request is wrong you send it back without disturbing the rest, and when you want to change direction you write a new plan. The repository keeps the complete story of how the game was built, and anyone can read it.