ai-sdk
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFE
Full Analysis
- [Indirect Prompt Injection] (SAFE): The skill provides patterns for processing untrusted user input within LLM workflows. 1. Ingestion points: Message arrays are ingested via API routes and client-side hooks. 2. Boundary markers: Delimiters are not explicitly shown in code snippets but are assumed to be handled by the SDK's message structure. 3. Capability inventory: The examples include tool execution (e.g., database writes, payments) and network-based LLM generation. 4. Sanitization: Tool inputs are strictly validated using Zod schemas, though raw text content is not sanitized before being passed to the model.
- [EXTERNAL_DOWNLOADS] (SAFE): The documentation refers to trusted packages from the Vercel ecosystem (ai, zod) and established ORMs (prisma, drizzle). No untrusted remote scripts or piped installations are present.
Audit Metadata