actualizar-imagen

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

• Third-party content exposure detected (high risk: 1.00). SKILL.md Fase 1 explicitly instructs the agent to use WebSearch/WebFetch to scrape public third‑party sources (e.g., Meta Ad Library, Dribbble, Pinterest, blogs and social media) with a prompt to extract and interpret ad content, and those parsed results are then used to update the repo and drive classification/changes—so untrusted external content can materially influence actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly calls WebFetch at runtime to pull content from external sites (e.g., https://foreplay.co and https://propphy.com) using an "Extract ALL..." prompt and then injects that fetched content into the model/output to generate or update image prompts and repo entries, so the external content directly controls prompts.