rag
Pass
Audited by Gen Agent Trust Hub on Aug 22, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is designed to build systems that process external, untrusted data (Retrieval-Augmented Generation). While it includes best practices like boundary markers and instructions to 'Answer based ONLY on the provided context,' the architectural pattern inherently creates an attack surface for indirect prompt injection.
- Ingestion points: The
RAGPipeline.answermethod inSKILL.mdingests query-relevant documents from aVectorStoreinto the LLM context. - Boundary markers: The implementation uses a system prompt instruction (
Answer based ONLY on the provided context) and separators (\n\n---\n\n) to delimit retrieved chunks. - Capability inventory: The provided Python snippets demonstrate LLM generation and vector search; they do not include dangerous subprocess or file-write capabilities.
- Sanitization: There is no explicit sanitization or filtering of the retrieved content before it is interpolated into the prompt template.
Audit Metadata