research

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

• Third-party content exposure detected (high risk: 1.00). The skill's required pipeline (Steps 3–5: "SEARCH" and "READ") explicitly runs searches and uses firecrawl/mcp/tavily to scrape 10–15 public web sources — including community sites like HN/Reddit/GitHub discussions — and then mandates the agent read and synthesize those per-source summaries into the report, so untrusted third-party content can directly influence downstream actions.