The Agent Skills Directory

[SAFE]: The skill does not exhibit any malicious behavior or security vulnerabilities. It follows appropriate patterns for API interaction and credential handling.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for pulling and running a Docker container from the official NVIDIA registry (nvcr.io/nim/mit/diffdock:2.2.0). This is a legitimate requirement for the skill's local execution mode and targets a trusted vendor.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied protein structures and ligand data which could theoretically serve as an injection vector. However, the risk is mitigated by the skill's specific scientific focus and the nature of the downstream API.
Ingestion points: Reads receptor data from protein.pdb and ligand data from ligand.sdf or user-provided SMILES strings.
Boundary markers: No explicit boundary markers are implemented for the interpolated data in the request payload.
Capability inventory: Uses Bash for environment setup, requests.post for network communication with NVIDIA endpoints, and local file writing for saving result poses.
Sanitization: Filters input PDB files to retain only ATOM records, effectively stripping potentially problematic metadata or non-standard records before transmission.

diffdock-nim