hud-evalshud-evalsOfficialยท1 Agent Skills Included

hud-python

Build, evaluate, and train models on RL environments

Builds reinforcement learning environments and tasks for coding, browser, computer-use, and robotics models. Runs graded evaluations across any model and turns every rollout into ready training data. Eliminates manual benchmark wiring, fragile grading scripts, and scattered evaluation infrastructure. Includes quality checks that catch reward hacking, contaminated benchmarks, and weak task designs before training.
npx skills add hud-evals/hud-python --all -g -y
Available:

Gives your AI agent the repository map, coding standards, testing rules, and debugging paths it needs to safely modify and extend the HUD evaluation SDK.

All Skills in This Repository (1)

Pure Emerald Level Indicators

Frequently Asked Questions

FAQPage Schema
How to install hud-python?โ–ผ

Run `npx skills add hud-evals/hud-python --all -g -y` in your terminal to install all skills in this suite globally.

What is HUD used for?โ–ผ

HUD lets you build RL environments, write graded tasks, and run evaluations or training for coding, browser, computer-use, and robotics models.

Which models work with HUD?โ–ผ

HUD natively supports Claude, OpenAI, Gemini, and OpenAI-compatible endpoints, and you can plug in your own custom harness or self-hosted model.

Can HUD train models with reinforcement learning?โ–ผ

Yes. Every graded rollout returns a reward and trace, which you can feed directly into GRPO or PPO training loops via the TrainingClient.

Do I need Docker to run HUD evaluations?โ–ผ

No. You can run `hud eval env.py model` locally with no Docker or cloud account, and only use containers when deploying environments at scale.

Related Repositories in Software Engineering

View All in Software Engineeringโ†’