assist

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md explicitly requires the planner agent to gather "業界/學術方案調研 + 社群共識/反面意見" and the HANDOFF template names public, user-generated sources ("GitHub discussions、SO、Reddit"), meaning the agent is expected to fetch and interpret open web/community content that can influence routing and actions.