agent-validation-v430 - Effective Agent Training with Safety Controls
Train agents to act effectively by disabling harmful actions and implementing tiered safety gates
Tags
Updated: 2026-05-09Capabilities
Typical Inputs
Typical Outputs
What this skill does
- disable entropy adjustment
- save best checkpoints
- consult agents
- apply actions
- adjust reward weights
- set tiered gates
- inject cross-run learning
- rollback model
- halt training
Inputs
- checkpoint path
- training environment
- agent metrics
Outputs
- model checkpoint
- training state
- reward weight update
- termination signal
Requirements
- Python 3.10
- PPO reinforcement learning
- GPU environment