service-innovation-case-study

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

• Third-party content exposure detected (high risk: 0.90). The skill's required workflow (e.g., "階段 1：基礎資料研究" and references/01-research-protocol.md) explicitly instructs querying and ingesting open public third‑party sources such as Crunchbase, TechCrunch, LinkedIn, App Store/Google Play reviews, Reddit and other media/user reviews and then uses those findings to drive PESTEL → SWOT → STP → strategy decisions, so untrusted user-generated content is read and can materially influence the agent's analysis and next actions.