article-writer

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

• Third-party content exposure detected (high risk: 1.00). The skill's research workflow (see rules/research.md, SKILL.md, and scripts/research.sh) explicitly performs web searches, URL/PDF crawling and "WebFetch" page fetching and ingests sources including user-generated sites (Reddit, Hacker News, Twitter/即刻/知乎), and those fetched pages are then used by the AI to generate drafts and drive subsequent actions—exposing the agent to untrusted third‑party content that can influence tool use.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill's research workflow explicitly accepts and automatically crawls user-supplied URLs at runtime (e.g., "https://vuejs.org/guide/"), using WebFetch to ingest remote page content into the knowledge base that will directly influence the agent's prompts and generated outputs, which meets the criteria for a runtime external dependency that can control agent behavior.