rag-implementation

Pass

Audited by Gen Agent Trust Hub on Apr 12, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill provides implementation patterns for Retrieval-Augmented Generation (RAG) which possess an inherent surface for indirect prompt injection.\n
  • Ingestion points: Untrusted document content is retrieved from external vector stores (e.g., Pinecone, Weaviate) in the code examples within SKILL.md.\n
  • Boundary markers: The prompt templates use text-based delimiters such as 'Context:' and 'Question:' but lack more robust isolation mechanisms.\n
  • Capability inventory: The system performs network-based API calls to LLM services (Anthropic) and vector database providers.\n
  • Sanitization: The provided implementation examples do not demonstrate input sanitization or content validation for the retrieved data before it is presented to the LLM.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 12, 2026, 09:37 AM
Security Audit — agent-trust-hub — rag-implementation