research

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

• Third-party content exposure detected (high risk: 1.00). The skill's core workflow (SKILL.md "execute" + "synthesize" / "deliver" steps) explicitly calls external backends (e.g., firecrawl, brave, exa, grok, last30days) that fetch/scrape public web pages and social media (firecrawl pages, Reddit/X in last30days, Grok/X) and requires the agent to read, synthesize, and cite those third‑party sources, so untrusted user-generated content can materially influence decisions and actions.