game-visual

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

• Third-party content exposure detected (high risk: 0.70). The required workflow explicitly instructs the agent to “search project-local references first, then use an allowed browser source only to fill a real gap” and to “archive each chosen external image or frame locally with source URL,” meaning runtime-fetched public web content (outsider-authored free text/prose around images) can be ingested into the LLM context via the browser-source step.