guardrails-developer-guide

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

• Third-party content exposure detected (high risk: 0.75). The required workflow fetches public documentation at runtime from https://docs.nvidia.com/nemo/guardrails/ (via llms.txt and .md page URLs), so outsider-authored free text from the public web is ingested into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill instructs the agent at runtime to fetch external documentation that will be loaded into the model context (controlling prompts), specifically https://docs.nvidia.com/nemo/guardrails/ and https://docs.nvidia.com/nemo/guardrails/llms.txt which are required for retrieving Markdown pages used as agent context.