The Agent Skills Directory

[SAFE] (SAFE): The file is a purely informational troubleshooting guide containing documentation and code examples.
Evidence: Code blocks consist of standard command-line flags for machine learning engines (SGLang) and training scripts.
Evidence: Python snippets demonstrate benign data validation, loss masking logic, and unit testing for reward functions using standard libraries like json and transformers.
Evidence: Environment variables such as NCCL_DEBUG and NCCL_TIMEOUT are standard configurations for distributed training environments.
Evidence: No evidence of obfuscation, remote code execution, credential exfiltration, or unauthorized persistence mechanisms was found.

slime-rl-training