The Agent Skills Directory

[SAFE]: The skill is a standard development and benchmarking tool provided by NVIDIA. Its operations are consistent with its stated purpose of evaluating RAG (Retrieval-Augmented Generation) quality.
[COMMAND_EXECUTION]: The skill provides commands to execute a local evaluation script scripts/eval/evaluate_rag.py using the uv package manager. These commands are necessary for the skill's functionality and target internal repository paths.
[EXTERNAL_DOWNLOADS]: The skill uses uv sync to manage Python dependencies for the evaluation script. These dependencies are standard for the task and are loaded through conventional package management workflows.
[CREDENTIALS_UNSAFE]: The skill references the use of an NVIDIA_API_KEY for RAGAS evaluation. It includes a dedicated 'Credential hygiene' section that correctly advises users against hardcoding secrets, recommending the use of environment files and secrets managers instead.
[DATA_EXFILTRATION]: Network operations are restricted to communication with a RAG server (defaulting to localhost) and an official NVIDIA API for judge scoring. There are no signs of unauthorized data exfiltration.

rag-eval