langchain

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

• Third-party content exposure detected (high risk: 0.75). Outsider free text can be ingested via the RAG runtime path where WebBaseLoader(url).load() fetches public web pages (arbitrary URL content) and the resulting docs/page_content are then embedded and passed into the LLM context through RetrievalQA(..., return_source_documents=True) / agent retriever tools.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill uses WebBaseLoader to fetch external webpages at runtime (e.g., https://docs.python.org/3/tutorial/), and those fetched documents are injected into the model context for RAG, meaning remote content directly controls the prompts/inputs the LLM sees.