semantius-agent-maker

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

• Third-party content exposure detected (high risk: 0.80). The maker explicitly instructs generated skills to perform "web research" against public sites and use those findings to dedupe/add catalog entries and drive writes (see the Frontmatter example "Pulls web research, dedupes against the existing catalog..." and "Pattern 1: Research-and-record" step 3: "Web research for candidates not in the baseline"), which means untrusted third‑party page content can be ingested and materially influence the agent's subsequent decisions and tool actions.