ai-news-collector

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md workflow explicitly instructs the agent to perform web searches and fetch/read public third-party content (e.g., "发现周报后，用 web_fetch 获取全文" and searches targeting site:reddit.com, news.ycombinator.com, Twitter/X, GitHub trending as listed in references/sources.md), so untrusted user-generated content will be ingested and used to drive news selection and hotness-ranking decisions.