The Agent Skills Directory

[SAFE]: No security issues detected. The skill provides defensive coding practices for AI agent development, specifically advising against the use of untrusted system roles in chat messages.- [EXTERNAL_DOWNLOADS]: References official documentation and GitHub repositories from Vercel, which is a recognized trusted organization. These references are used for standard SDK implementation guidance.- [INDIRECT_PROMPT_INJECTION]: The skill demonstrates secure patterns for handling untrusted user input by using server-side instructions and strict Zod schema validation for tool inputs. Ingestion points: The messages parameter in SDK calls within references/request-handlers.md represents the ingestion of external data. Boundary markers: The SKILL.md file explicitly mandates rejecting untrusted { role: 'system' } messages and setting allowSystemInMessages: false. Capability inventory: The tools provided in templates/agent-with-tools.ts are limited to benign mock data and mathematical operations. Sanitization: Tool inputs are strictly validated using Zod schemas as demonstrated in templates/agent-with-tools.ts.

ai-sdk-core-v7