gauss-atropos-environments

Build, test, and debug Gauss RL environments for Atropos training.

1.2k|116|Updated Mar 19, 2026
One-click install
npx skills add https://github.com/math-inc/OpenGauss --skill gauss-atropos-environments
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: gauss-atropos-environments
Source: https://github.com/math-inc/OpenGauss/tree/main/skills/mlops/training/gauss-atropos-environments
Command: npx skills add https://github.com/math-inc/OpenGauss --skill gauss-atropos-environments

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Build, test, and debug Gauss RL environments for Atropos training.

Core Features & Use Cases

  • Multi-turn agent loop with tool calling integration for Gauss environments.
  • Provides GaussAgentBaseEnv interface, tool context, evaluation with full agent loop, wandb logging, and CLI modes serve/process/evaluate.
  • Use case: rapid development and benchmarking of RL environments in gauss-agent repo.

Quick Start

Create your own GaussAtropos environment by subclassing GaussAgentBaseEnv and running the provided CLI using the serve, process, or evaluate modes.

Frequently Asked Questions about gauss-atropos-environments

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I build an RL environment for Atropos training with multi-turn agent loops?

Build an RL environment for Atropos by subclassing the GaussAgentBaseEnv interface to implement multi-turn agent loops with tool calling, structured reward computation, and evaluation workflows. This provides the core interface for rapid development and benchmarking.

What is the best way to test and debug Gauss RL environments locally?

The best way to test and debug Gauss RL environments is using the three CLI modes: serve for running the environment, process for data generation, and evaluate for sandboxed evaluation with full agent loops. These modes enforce reproducible workflows during development.

Does this environment interface support tool calling and wandb logging integration?

Yes, the environment interface supports tool calling integration within the multi-turn agent loop and wandb logging for tracking metrics. Tool context is provided natively, and evaluation loops run with full agent loop capabilities for comprehensive benchmarking.

How does structured reward computation work with AgentResult data in Gauss environments?

Structured reward computation uses AgentResult data to calculate rewards in Gauss environments during the evaluation phase. This ensures reproducible workflows and sandboxed evaluation by enforcing consistent data structures throughout the agent loop process.

Can I run evaluation loops with full tool integration in a sandboxed environment?

Yes, you can run evaluation loops with full tool integration in a sandboxed environment using the evaluate CLI mode. This enforces reproducible workflows while testing the multi-turn agent loop and structured reward computation with AgentResult data.