rag-implementation
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a RAG (Retrieval-Augmented Generation) architecture which creates an attack surface for indirect prompt injection if the ingested documents contain malicious instructions intended to subvert the agent's behavior.
- Ingestion points: External data enters the agent context through vector database retrieval methods (e.g.,
vectorstore.as_retriever,parent_retriever.ainvoke) as shown in the Python code blocks withinSKILL.md. - Boundary markers: Prompt templates in the
generateandHyDEnodes use simple headers like 'Context:' and 'Question:' but do not implement robust delimiters or explicit 'ignore instructions' warnings for the retrieved content. - Capability inventory: The system performs network operations to vector databases and executes LLM generation cycles based on retrieved data.
- Sanitization: The provided examples do not demonstrate sanitization, filtering, or validation of the retrieved document content before it is interpolated into the final LLM prompt.
Audit Metadata