ai-provider-claude-vision

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

• Third-party content exposure detected (high risk: 0.80). This skill explicitly supports and instructs sending images and PDFs via public URL content blocks (see "Pattern 2: URL vs Base64 vs Files API" in SKILL.md and the "URL Image Input" / "PDF Document Input (URL)" examples), meaning the agent will fetch and directly read untrusted third-party web-hosted content that could contain instructions influencing its actions.