deep-research

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

• Third-party content exposure detected (high risk: 0.90). This skill explicitly launches a web-search-agent (see agents/web-search-agent.md and the /research-deep and /research workflows in research-deep/SKILL.md and research/SKILL.md) that searches and ingests open public, user-generated sources (e.g., Reddit, Stack Overflow, GitHub, X/Twitter, blogs, arXiv) and then uses those results to populate JSON outputs and reports, so untrusted third‑party content can materially influence agent decisions and actions.