cali-degustia-diagnostico

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly tells the LLM to "read" external guidance pages (e.g., sureprompts.com/blog/chatgpt-custom-gpts-guide and platform.claude.com/... plus claude.com/...) during runtime to learn templates/best-practices, meaning those URLs would be fetched at execution time and can directly shape prompts/instructions.