whox-atropos-environments

What problem does it solve?

Build, test, and debug WHOX Agent RL environments for Atropos training. It covers the WHOXAgentBaseEnv interface, reward functions, agent loop integration, evaluation with tools, wandb logging, and the three CLI modes (serve, process, evaluate). Use this skill when creating, reviewing, or fixing RL environments in the whox-agent repo.

Core Features & Use Cases

• Environment scaffolding: provides a structured pattern for implementing a base WHOX environment that integrates with Atropos and the WHOXAgentBaseEnv, including setup, item queuing, evaluation loops, and tool usage.
• Agent-loop evaluation: demonstrates running a full agent loop with tool calls for end-to-end assessment rather than single-turn completions.
• Wandb logging and metrics: includes hooks for performance tracking, custom rewards, and evaluation logging to assist reproducibility and benchmarking.
• CLI modes: supports serve, process, and evaluate workflows to enable rapid development, offline data generation, and model benchmarking.

Quick Start

Implement a WHOXAgentBaseEnv-based environment and run the provided CLI in serve or evaluate mode to validate end-to-end RL workflows.