nemo-retriever-mcp
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for ingesting and querying external documents, which represents an indirect prompt injection surface.
- Ingestion points: Data enters the agent's context through the
queryandingest_documentstools as described inSKILL.md. - Boundary markers: The instructions include a requirement to "ground your answer" in evidence hits, which provides some context separation, but the skill lacks explicit delimiters or instructions to ignore commands embedded within retrieved document text.
- Capability inventory: The skill uses tools to read and add documents to an index, which are then processed by the LLM to generate answers.
- Sanitization: No sanitization, filtering, or validation of the retrieved document content is described in the skill instructions.
- [NO_CODE]: The skill package contains no executable scripts (Python, JavaScript, shell), binaries, or installation files; it consists solely of markdown and JSON documentation.
- [SAFE]: The skill uses official documentation links from the vendor's own domain (nvidia.com) and provides legitimate technical instructions for using the specified tools without attempting to bypass safety filters or exfiltrate data.
Audit Metadata