This repo is for getting hands-on with AI agent environments: running existing ones, reproducing their results, and seeing where they break.
First up is a small existing environment from Prime Intellect's verifiers framework. The plan is to run it unchanged and record:
- environment and framework version
- model and configuration
- exact commands
- trajectory and score
- anything that behaves differently from the documentation
Files from the run will live here:
reproductions/verifiers-example/
├── README.md
├── commands.sh
├── environment.txt
└── results/
└── baseline.json