The Agent Skills Directory

[SAFE] (SAFE): The file is a technical reference guide and does not contain any executable malicious code. All code snippets provided are illustrative examples of standard AI SDK usage.
[Indirect Prompt Injection] (SAFE): The documentation describes an agentic framework that naturally processes untrusted input. This is the primary intended purpose of the technology.
Ingestion points: User input via the prompt parameter in generate and stream methods, and external context retrieved in RAG patterns.
Boundary markers: Not explicitly detailed in the brief snippets, which is expected for documentation.
Capability inventory: The framework supports tool execution via defined execute functions.
Sanitization: The reference focuses on SDK implementation; input sanitization is the responsibility of the application developer.

vercel-ai-sdk