ai-sdk

Pass

Audited by Gen Agent Trust Hub on Sep 1, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions for installing official Vercel AI SDK packages (e.g., ai, @ai-sdk/openai, @ai-sdk/react) from the npm registry. These are well-known, trusted dependencies for TypeScript development.
  • [INDIRECT_PROMPT_INJECTION]: Documentation includes examples of processing untrusted external data (such as articles or user-provided files). The skill identifies the injection surface and mitigates it by explicitly warning against using allowSystemInMessages with untrusted history and documenting best practices for toolApproval to maintain human-in-the-loop control.
  • [DYNAMIC_EXECUTION]: The reference material describes the SDK's ability to execute model-generated code via sandboxed runtimes like QuickJS or Vercel Sandbox. These features are documented as standard SDK capabilities for build-time and runtime agent tasks.
  • [COMMAND_EXECUTION]: Mentions of terminal UI and sandbox execution tools (e.g., @ai-sdk/tui, @ai-sdk/sandbox-just-bash) are provided in the context of developer tooling and sandboxed agent environments, following the SDK's intended usage patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 1, 2026, 04:34 PM
Security Audit — agent-trust-hub — ai-sdk