ai-sdk

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime workflow only instructs the agent to search bundled local docs/source in node_modules/ai/ (or to fetch specific .md URLs from ai-sdk.dev), and the only LLM-readable text ingestion path described is fetching those .md pages from a public website (outsider-authored web content) when using the “Otherwise” branch.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs runtime fetching from https://ai-sdk.dev/api/search-docs?q=your_query and direct retrieval of .md URLs (e.g. https://ai-sdk.dev/docs/agents/building-agents.md), which would inject external documentation into the agent’s context and thus can directly control prompts at runtime.