consulting-problem-solving-ru

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly delegates internet research to a general-purpose subagent (SKILL.md: "делегируй сабагенту через инструмент Task (subagent_type: "general-purpose")") and Step 5 / citation rules (references/05-analyze.md and SKILL.md) require ingesting and using public third‑party sources (e.g., Bureau of Labor Statistics, Росстат, McKinsey) as evidence that will materially influence analyses, decisions, and subsequent tool-driven actions, so untrusted web content could introduce indirect prompt injection.