speech-act-pragmatic

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime LLM context is driven by the corpus text units being classified (Step 2/Step 3), and if those text units include outsider-authored free-form content (e.g., scraped web pages, comments, issues, or other users’ messages), that prose is ingested into the agent/LLM for classification and reporting—an indirect prompt-injection risk.