hermes-atropos-environments

To build Atropos RL environments, you implement the HermesAgentBaseEnv interface by defining setup, get_next_item, evaluate, compute_reward, and format_prompt methods to enforce the base environment contract.

Hermes Agent RL environments use a defined tool-resolution workflow and sandboxed evaluation to validate tool interactions, ensuring tool calls function correctly before computing reward signals during training.

You can debug reward functions in an Atropos agent loop by running a quick deterministic rollout, which validates tool integration and sandboxed evaluation to isolate and troubleshoot reward signals.

Yes, Hermes Agent Atropos environment development includes architecture-aware scaffolding for wandb logging to track evaluation loops and reward computation metrics during training.

A Hermes Agent base environment requires defining setup, item retrieval, reward computation, evaluation loops, and wandb logging, while strictly enforcing the Atropos base environment contract and required config.

Atropos RL environment evaluation requires using specific documented CLI modes to run deterministic rollouts, validate tool interactions, and test reward signals within the HermesAgentBaseEnv interface.