rfd4-proteina-nim
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [SAFE]: The skill uses established security patterns for interacting with external services. Path traversal is prevented by strict filename validation in scripts/design.py. Credential exposure is mitigated by disabling HTTP redirects and restricting authentication headers to specific vendor endpoints.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes biochemical data files and API responses which enter the agent context.\n
- Ingestion points: mmCIF files in assets/requests/ and JSON responses from the NIM inference service.\n
- Boundary markers: Structured output files (transport.json, summary.json) are used to manage results and separate model metadata from design data.\n
- Capability inventory: Regulated bash execution of the design.py script, with file system access (Read/Write) and network fetching capabilities.\n
- Sanitization: The design.py script validates inputs against a strict JSON schema and checks the structural integrity and coordinate finiteness of returned design data.\n- [COMMAND_EXECUTION]: Shell access is strictly limited to the scripts/design.py helper script via YAML frontmatter, preventing arbitrary command injection by the agent.\n- [EXTERNAL_DOWNLOADS]: The skill documentation describes how to deploy the NVIDIA RFD4-Proteina NIM container from the official NVIDIA NGC registry (nvcr.io).
Audit Metadata