ecosystem-primer

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

• Third-party content exposure detected (medium risk: 0.65). The required workflow includes Step 3 “How the Docs Work” where, if an MCP docs server isn’t available, it fetches live public web content (e.g., https://docs.langchain.com/llms.txt and then specific docs pages) and injects that page text into the agent context for answering questions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill instructs at runtime to fetch external documentation to drive agent behavior (identify pages and fetch them), specifically calling https://docs.langchain.com/llms.txt which would directly influence prompts/instructions.