Evaluate and improve models and agents using environments
NVIDIA-NeMo/Gym is a library and infrastructure for evaluating and improving models and agents inside environments, where each environment defines tasks, agent interaction, verification, and execution state. It is for teams running reproducible evaluations or training at scale across settings such as code execution, tool calling, and sandboxes, and the catalogue entries provide skills and instructions for working with it.
Latest release v0.7.0 · 12 Oct 2026
17 files for Claude Code, Codex and OpenCode: add-benchmark, gh-stack, nemo-gym-blade-analysis, nemo-gym-debugging and 13 more — 5,156 tokens loaded in every session.
AGENTS.md A 5,153 tok CLAUDE.md A 3 tok .agents/skills/add-benchmark/SKILL.md A 120 tok .agents/skills/gh-stack/SKILL.md A 65 tok .agents/skills/nemo-gym-blade-analysis/SKILL.md A 106 tok .agents/skills/nemo-gym-debugging/SKILL.md A 68 tok .agents/skills/nemo-gym-docs/SKILL.md A 80 tok .agents/skills/nemo-gym-pivot-datasets/SKILL.md A 97 tok .agents/skills/nemo-gym-reward-profiling/SKILL.md A 78 tok .agents/skills/pr-review/SKILL.md A 22 tok .agents/skills/nemo-gym-pr-checks-and-labels/SKILL.md A 123 tok .agents/skills/nemo-gym-release-notes/SKILL.md A 50 tok .agents/skills/nemo-gym-release-publish/SKILL.md A 108 tok .agents/skills/add-harness/SKILL.md A 172 tok .agents/skills/nemo-gym-score-triage/SKILL.md A 133 tok .agents/skills/add-model-eval-config/SKILL.md A 188 tok .agents/skills/run-and-babysit-eval/SKILL.md C 190 tok These files are NVIDIA-NeMo/Gym's own configuration — they tell Claude Code, Codex and OpenCode how to work on this repository, so they are not mods to install elsewhere. Copy one as a starting point and replace the parts that are about this project.