The Agent Skills Directory

[SAFE]: The skill content provides legitimate guidance for nvmolkit, a library for GPU-accelerated cheminformatics operations developed by the vendor nvidia-bionemo.
[EXTERNAL_DOWNLOADS]: The skill references official documentation and installation sources that are consistent with the vendor identity. Specifically, it points to nvidia-bionemo.github.io and github.com/NVIDIA-BioNeMo. It also correctly directs users to well-known services such as pytorch.org and standard package registries like conda-forge.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-supplied molecular datasets and SMILES strings.
Ingestion points: Data enters the system via SMILES parsing in SKILL.md and references/advanced-usage.md.
Boundary markers: The instructions explicitly advise to "Parse SMILES with RDKit and reject failed parses," establishing a validation checkpoint for external data.
Capability inventory: The skill uses standard file system operations for results (Path.write_text) and GPU tensor manipulation through torch.
Sanitization: The use of RDKit's molecule parser acts as a robust sanitization layer for molecular inputs.
[METADATA_POISONING]: The benchmark documentation notes a minor linting observation regarding the author field format, which is a best-practice violation but does not impact security.

nvmolkit-usage