landing-pages

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md registry explicitly lists external landing page URLs and states "Claude will analyze each page on first use" and can "re-fetch" those URLs (e.g., https://example.com/pricing), meaning the agent fetches and interprets arbitrary public web pages which could contain untrusted, user-controlled content that influences ad generation.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly fetches and analyzes external landing pages at runtime (e.g., https://example.com/pricing and other https://example.com/* variants), injecting that content into Claude's analysis/context to drive ad-generation prompts, so the external URLs directly control the agent's prompts.