The Agent Skills Directory

[SAFE]: No security issues detected. The skill consists of technical documentation and scripts for distributed training workflows.\n- [COMMAND_EXECUTION]: The skill utilizes standard commands for the Isaac Lab environment, such as uv run isaaclab train_multigpu, to launch training tasks. It also uses nvidia-smi to inspect system topology, which is standard practice for distributed ML.\n- [DYNAMIC_EXECUTION]: A standalone Python probe (nccl_probe.py) is provided to diagnose NCCL hangs. The script performs a simple all_reduce operation using the torch.distributed library to verify that inter-GPU communication is functional.

isaaclab-training-multi-gpu