regex-vs-llm-structured-text

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

• Third-party content exposure detected (high risk: 0.70). The required workflow feeds the runtime “Source Text” (content) into the LLM via validate_with_llm() when confidence is low, and that source text is not guaranteed to be authored by the operating user (it could be outsider-provided document text), creating an indirect prompt-injection path through messages.create(... "Text: {original_text}").