langchain

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md RAG/agent example explicitly uses WebBaseLoader(web_paths=(...)) to load web pages (e.g., loader.load() on public URLs) and a retrieve_context tool that returns serialized retrieved_docs for the agent to read and use in answering, meaning untrusted third-party web content can be ingested and directly influence agent actions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's runtime code uses WebBaseLoader to fetch content from https://example.com/docs and injects those documents into the retrieval/context for RAG prompts, meaning external content directly controls the agent's prompt/context.