van-tu-tu-training

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

• Third-party content exposure detected (high risk: 0.80). The skill's Step 1 "Quy tắc làm giàu thông tin" explicitly instructs the agent to "chủ động dùng WebSearch hoặc skill deep-research" to fetch benchmarks, case studies and trends from the open web when the user's topic is vague, and to merge those third‑party insights into the required Training Brief—meaning untrusted public content is fetched, interpreted as part of the workflow, and can materially influence subsequent design decisions.