The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides code patterns for building chat interfaces using the Vercel AI SDK. These patterns involve passing user-provided messages directly to a language model, which is a common implementation but constitutes an attack surface for indirect prompt injection.
Ingestion points: The POST route in app/api/chat/route.ts extracts messages directly from the request body.
Boundary markers: The implementation uses a basic system prompt without explicit delimiters or warnings to the model to ignore potential instructions embedded in user content.
Capability inventory: The untrusted data is processed by the streamText capability to generate AI responses.
Sanitization: No specific input sanitization or filtering logic is present in the provided architectural examples.
[SAFE]: The skill documents best practices for environment variable management, specifically recommending the use of process.env and validation to avoid hardcoding credentials. It references standard, well-known libraries and services appropriate for the described use case.

vercel-development