The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions and shell command templates for deploying a local NVIDIA NIM using Docker.
Evidence: SKILL.md and references/api.md include commands for docker login, docker run, and polling a local health check endpoint (/v1/health/ready).
Security Posture: The skill includes security-conscious configurations, such as binding the local API to 127.0.0.1 and advising the agent to disable shell tracing (set +x) to protect credentials during setup.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading a container image from NVIDIA's registry and installing a Python package for chemical notation conversion.
Evidence: Mentions docker login nvcr.io for pulling nvcr.io/nim/nvidia/genmol:1.0.1 and recommends pip install safe-mol in references/api.md.
Source Analysis: Both nvcr.io and the safe-mol package are official resources provided by the skill author (NVIDIA).
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests and processes user-provided molecular data (SMILES/SAFE strings).
Ingestion Points: User-supplied chemical scaffolds and motifs are interpolated into API request payloads and potentially processed by the safe-mol library.
Boundary Markers: The skill utilizes JSON payloads for API communication, which provides inherent structure, but does not explicitly define delimiters for untrusted chemical notation strings.
Capability Inventory: The skill utilizes Bash for container management and requests for network operations to hosted or local endpoints.
Sanitization: The skill mentions handling SAFEFragmentationError and recommends validation using RDKit in references/validation.md to ensure chemical validity.

genmol-nim