ai-orchestration-vercel-ai-sdk
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from user prompts and external tool outputs (e.g., web search results, weather data) by interpolating them directly into LLM prompts, creating a potential surface for indirect instruction execution.
- Ingestion points: External data enters the agent context via variables such as
articleinexamples/core.md,topicinexamples/structured-output.md, andrelevantChunksinexamples/rag.md. - Boundary markers: Most examples lack robust delimiters or explicit instructions to ignore embedded commands, often relying on simple labels like
Context:orArticle:. - Capability inventory: The skill facilitates network operations through the
fetchAPI in tool definitions (e.g.,examples/tools.md) and manages external tool execution flows. - Sanitization: No explicit sanitization or filtering of external content is shown before it is interpolated into model prompts.
- [EXTERNAL_DOWNLOADS]: The skill documentation recommends installing standard packages from the Vercel AI SDK ecosystem and the Zod validation library from public registries.
- Evidence: References to
ai,@ai-sdk/openai,@ai-sdk/anthropic,@ai-sdk/google,@ai-sdk/react, andzodare found throughoutreference.mdand various code examples.
Audit Metadata