Verifiers is a Python library for creating environments that train and evaluate large language models. An environment defines the task a model is asked to perform and the checks that score what it produces, so the same definition serves both evaluation runs and reinforcement-learning training.
It is used with the Environments Hub, where environments are published and installed from, the prime-rl training framework, and a hosted training platform from the same project. The Prime command-line tool, installed with uv, is the recommended way to work with published environments. Architecture guides sit in the repository's docs directory, with an AGENTS.md file and a skills directory covering the same ground in more detail for coding agents.
Features
- Environment authoring: define the task and the verification that scores a model's output
- Environments Hub: environments are published to and installed from a shared hub
- Training integration: works with the prime-rl training framework and a hosted training platform
- CLI access: the Prime command-line tool installs and runs published environments
- Agent-oriented docs: an AGENTS.md file and a skills directory document the internals for coding agents
