The Agent Skills Directory

[SAFE]: The skill serves as a technical reference guide for LLaMA-Factory, aggregating official documentation to assist users in model fine-tuning tasks.
[EXTERNAL_DOWNLOADS]: References official project documentation at llamafactory.readthedocs.io and instructs on downloading models from well-known repositories like Hugging Face and ModelScope. These activities represent standard operations within the machine learning ecosystem.
[COMMAND_EXECUTION]: Includes standard command-line instructions for launching training processes via llamafactory-cli, torchrun, deepspeed, and accelerate. These are intended for local execution by the user to manage training jobs.
[REMOTE_CODE_EXECUTION]: Mentions the trust_remote_code parameter, which is a legitimate and common feature of the Transformers library for loading custom model architectures from remote hubs. The skill treats this as a standard configuration option.
[CREDENTIALS_UNSAFE]: Documentation specifies parameters for authentication (such as hf_hub_token). No actual credentials are hardcoded; the skill correctly identifies these as user-provided arguments for accessing private Hub resources.

llama-factory