The Agent Skills Directory

[SAFE]: The skill follows security best practices by explicitly advising against the disclosure of secret values like NVIDIA_API_KEY in the chat and recommending the use of environment variables or local configuration files.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute nemotron CLI commands for retrieval recipe management. The instructions favor read-only checks, help commands, and dry-runs to ensure command correctness before execution.
[EXTERNAL_DOWNLOADS]: The skill references models and datasets from trusted sources, specifically official NVIDIA repositories on HuggingFace and the NVIDIA Container Registry. These downloads are standard for the skill's intended synthetic data generation and deployment tasks.
[PROMPT_INJECTION]: The skill processes external text corpora for training purposes. This represents an indirect prompt injection surface; however, the risk is minimized as the data is used for offline model training rather than direct instruction interpolation, and the skill advises users to adhere to data-governance policies.
Ingestion points: Text corpora specified via file paths or HuggingFace URIs in the nemotron embed/rerank commands.
Boundary markers: Not applicable for the primary use case of model training.
Capability inventory: Execution of repo-scoped CLI tools for training, evaluation, and deployment stages.
Sanitization: Not explicitly performed on the raw corpus text within the instructions, relying instead on user policy compliance.

nemotron-retrieval-recipes