The Agent Skills Directory

[SAFE]: The skill instructions and examples are consistent with the documented purpose of NeMo AutoModel recipe development. The content focuses on legitimate configuration patterns and validation workflows.
[DYNAMIC_EXECUTION]: The skill documents the _target_ field in YAML configurations, which allows the framework to instantiate Python callables dynamically. This is a core architectural feature of the underlying Hydra-based configuration system used by NeMo to build components like models, optimizers, and datasets.
[INDIRECT_PROMPT_INJECTION]: The skill describes a workflow that ingests external configuration and data, which represents a potential attack surface. However, the risk is managed by the framework's architecture and the intended use case for developer-controlled training recipes.
Ingestion points: External YAML configuration files and dataset files (specified in the dataset section of the YAML anatomy).
Boundary markers: None explicitly defined in the skill body.
Capability inventory: The skill facilitates the execution of training and evaluation logic through the automodel CLI tool, which performs file operations and executes computation tasks via built modules.
Sanitization: Not explicitly documented within the instructions; the skill relies on the underlying NeMo framework's handling of configuration values.

nemo-automodel-recipe-development