deep-research

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

• Third-party content exposure detected (high risk: 0.90). The skill's SKILL.md and Parallel Search instructions explicitly require using WebSearch/WebFetch (Step 2 "Searches its axis using WebSearch and WebFetch") and even instruct sub-agents to download external PDFs with curl (references/parallel-search.md), and type-specific references (e.g., references/community.md) direct scraping of public forums/Reddit/Discord and other user-generated sources, so the agent will fetch and interpret untrusted third‑party content that can influence its analysis and actions.