langchain-architecture

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime workflow includes a RAG pattern where the agent retrieves Document objects from an external vector store/retriever (e.g., PineconeVectorStore.as_retriever()), and those documents’ page_content is concatenated into the LLM prompt (context_text = "\n\n".join(doc.page_content ...)), which can be outsider-authored content depending on what was indexed.