The Agent Skills Directory

[SAFE]: The skill provides technical documentation and instructions for the NVIDIA NeMo AutoModel framework. All external resources, including documentation links and GitHub repository references, belong to the official NVIDIA vendor namespace or recognized trusted organizations like Hugging Face.
[COMMAND_EXECUTION]: The skill instructions include examples of using the automodel command-line interface for local validation and training. These commands are project-specific tools intended for the stated purpose of recipe development and do not involve unauthorized or high-risk execution patterns.
[DYNAMIC_EXECUTION]: The skill documents the _target_ pattern used to specify Python callables within YAML configurations. This is a core architectural feature of the NeMo and Hydra frameworks used to instantiate components like optimizers and datasets from trusted libraries such as torch and nemo_automodel.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to help agents process YAML configurations. While ingesting external configuration files presents a theoretical surface for indirect injection, the skill operates within the expected domain of machine learning development and does not expose sensitive capabilities or bypass safety guardrails.

nemo-automodel-recipe-development