The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local protein structure files (e.g., target.pdb) and interpolates this content into API request payloads.
Ingestion points: SKILL.md and references/examples.md provide code patterns that read the content of target.pdb for submission to the RFDiffusion endpoint.
Boundary markers: The skill does not implement explicit delimiters or "ignore embedded instructions" warnings when interpolating PDB data into the JSON payload.
Capability inventory: The skill utilizes requests.post for network operations, Path().write_text() for file system writes, and supports shell command execution for Docker setup.
Sanitization: No sanitization or validation of the PDB content is performed before it is processed and sent to the remote service.
[SAFE]: The skill provides instructions for managing credentials (NGC_API_KEY) via environment variables and .env files, which aligns with standard security best practices for secret management.
[SAFE]: External resource references, including the Docker image hosted on nvcr.io and API endpoints at health.api.nvidia.com, are official resources provided by the vendor (NVIDIA) and are documented neutrally as intended for the primary function of the skill.

rfdiffusion-nim