vercel-ai-sdk

Pass

Audited by Gen Agent Trust Hub on Feb 24, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references documentation and source code from Vercel's official resources.
  • Evidence: Links to https://sdk.vercel.ai/docs and https://github.com/vercel/ai.
  • [PROMPT_INJECTION]: The skill implements patterns that interpolate untrusted user data directly into LLM prompts, creating an indirect prompt injection attack surface.
  • Ingestion points: User input is processed in SKILL.md (Pattern 3), references/examples.md (RecipeExtractor component), and references/streaming.md (CompletionDemo component).
  • Boundary markers: Examples provided show user input being embedded directly into task instructions without the use of clear delimiters or instructions to ignore embedded commands (e.g., `Extract recipe information: ${prompt}`).
  • Capability inventory: The skill demonstrates tool calling capabilities in references/streaming.md allowing the agent to perform external network requests (fetchWeather, searchAPI) based on model decisions.
  • Sanitization: While the skill emphasizes the use of Zod for structured output validation, it lacks examples or guidance on sanitizing the user input strings before they are incorporated into the prompt context.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 24, 2026, 05:18 PM