The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the use of standard Docker commands to deploy the local Evo 2 NIM microservice. This includes logging into the official NVIDIA container registry (nvcr.io) and running the container with appropriate GPU runtimes. These operations are essential for the skill's primary function.
[EXTERNAL_DOWNLOADS]: Network operations are restricted to well-known, trusted vendor infrastructure. The skill communicates with the official NVIDIA hosted API (health.api.nvidia.com) and pulls Docker images from the official NVIDIA registry (nvcr.io). No downloads from untrusted third-party sources were detected.
[DYNAMIC_EXECUTION]: The skill processes tensor data returned from the local forward pass endpoint using numpy.load. It correctly implements security mitigations by explicitly setting allow_pickle=False, which prevents potential code execution vulnerabilities associated with unsafe deserialization of NumPy arrays.
[CREDENTIALS_UNSAFE]: The skill uses environment variables (NGC_API_KEY and NVIDIA_API_KEY) for authentication to the hosted API and registry. It follows best practices by suggesting these be stored in a local .env file rather than being hardcoded. It also sets restrictive permissions (chmod 700) on the local cache directory to protect data.
[INDIRECT_PROMPT_INJECTION]: The skill implements a clean_dna validation function that sanitizes user-provided DNA sequences by ensuring they only contain valid nucleotide bases (A, C, G, T) before processing. This reduces the risk of data poisoning or unexpected behavior from malformed inputs.

evo2-nim