The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides shell commands for Docker orchestration, including authentication with docker login nvcr.io and service deployment via docker run. These commands are intended to facilitate the setup of the local inference environment.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external protein structure data (target.pdb), which represents a surface for indirect prompt injection.
Ingestion points: Data from target.pdb is ingested into the agent context in the motif scaffolding and binder design sections of SKILL.md and references/examples.md.
Boundary markers: The skill does not define explicit delimiters or provided warnings to the agent to disregard instructions potentially embedded within the ingested PDB content.
Capability inventory: The skill possesses Bash execution capabilities for Docker management and Write access to save results in SKILL.md, and performs network operations via the requests library in references/examples.md.
Sanitization: No structural validation or sanitization is performed on the ingested PDB content before it is included in API request payloads.

rfdiffusion-nim