krea-archviz

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly ingests and "Read"s user-supplied external images/URLs (e.g., "Receive the screenshot from the user (attached file, URL, or pasted image)" in references/screenshot-to-render.md and "Read EACH reference with vision" in references/composition.md), and those untrusted third‑party references are parsed and used to choose models and drive generation, so external content can materially influence agent actions.