langchain

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md explicitly shows loading and ingesting public web pages (e.g., RAG → Document Loading using WebBaseLoader("https://example.com")) and defines a web-search tool/agent (search_web and Agents sections) that ingest and use open-web content as context for LLMs, which are untrusted third‑party sources that can influence model behavior and enable indirect prompt injection.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill performs runtime pulls of external content that can control prompts—e.g., hub.pull("hwchase17/react") which fetches a prompt template used to construct the agent, and WebBaseLoader("https://example.com") which loads webpage content that can be injected into RAG prompts—so these external resources directly influence agent instructions at runtime.