The Agent Skills Directory

[SAFE]: The skill follows secure development practices by utilizing official, well-known libraries such as @ai-sdk/openai, @ai-sdk/anthropic, and zod. All external references point to official documentation and repositories from Vercel.
[INDIRECT_PROMPT_INJECTION]: The templates demonstrate standard patterns for building AI applications where user input is passed to models via generateText and streamText. While this presents a common attack surface for LLM applications, the skill correctly uses instructions (system prompts) to define model boundaries, which is the standard mitigation for this environment. No high-risk capabilities are exposed through the included tools.
[REMOTE_CODE_EXECUTION]: The skill does not contain any patterns for downloading or executing remote code. All logic is contained within the local templates, and tool execution is restricted to mock data and basic calculations.
[DATA_EXFILTRATION]: No credential hardcoding or exfiltration attempts were found. The instructions to read local node_modules documentation are restricted to specific, non-sensitive paths within the project directory to support the developer's workflow.

ai-sdk-core-v7