deep-research

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

• Third-party content exposure detected (high risk: 1.00). The skill's mandatory Phase 3 "RETRIEVE" workflow in SKILL.md explicitly launches parallel WebSearch / mcp__Exa__exa_search / WebFetch calls and spawns Task agents to collect and read public web/academic/industry sources (e.g., arXiv, news, arbitrary domains), and those ingested, untrusted third‑party contents are then used to triangulate, synthesize, and drive subsequent actions and report generation—creating a clear avenue for indirect prompt injection.