social-media-intelligence

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

• Third-party content exposure detected (high risk: 0.85). Outsider-authored free text from public social platforms (e.g., Twitter/X, Telegram channel messages, Discord channel history, Reddit post titles/selftext) is collected at runtime and then fed into the LLM via the llm_sentiment() workflow (prompt = ... Text: {text[:1000]} → get_llm().invoke(prompt)), creating an indirect prompt-injection surface.