marketing-content

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

• Third-party content exposure detected (medium risk: 0.30). In “Research Mode: Content Intelligence,” the workflow uses agent-browser to open and then ingest free-text from outsider-authored web sources (Google PAA results pages, AnswerThePublic, competitor blogs, Reddit search results, and Quora search pages) via “agent-browser get text body.”

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

• Potentially malicious external URL detected (high risk: 0.90). The skill instructs installing and using agent-browser at runtime and includes the command that fetches and executes remote code from https://github.com/vercel-labs/agent-browser (npx skills add ...), which is a required runtime installation that can execute remote code used by the agent-browser commands in SKILL.md.