The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads external PDB files provided by the user to process them for protein design. This ingestion represents a potential surface for indirect prompt injection.
Ingestion points: PDB file content is read into a string in scripts/design.py via pdb.read_text().
Boundary markers: The content is treated as a raw string within a JSON payload; no specific boundary markers for natural language instructions are present.
Capability inventory: The skill has the capability to perform network requests (requests.post) and write output files to the local system.
Sanitization: The script verifies the presence of ATOM records but does not perform sanitization for embedded natural language instructions.
[COMMAND_EXECUTION]: The skill provides a helper script scripts/design.py and reference Docker setup commands to facilitate protein design tasks.
scripts/design.py is used to send inference requests to the NIM API.
references/api.md provides setup documentation including docker run commands.
[EXTERNAL_DOWNLOADS]: The skill references Docker images from the official NVIDIA Container Registry (nvcr.io).
These references target a well-known and trusted vendor for the purpose of local service deployment.

proteinmpnn-nim