The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for building AI agents that process untrusted user input and interact with external tools. This architectural pattern represents an indirect prompt injection surface.
Ingestion points: User-supplied strings are interpolated into prompts via the prompt and messages parameters in templates such as templates/generate-text-basic.ts and templates/agent-with-tools.ts.
Boundary markers: The skill recommends using human-in-the-loop approval processes for sensitive tool execution in SKILL.md to limit the impact of malicious data.
Capability inventory: The skill demonstrates tool-calling capabilities including filesystem operations via MCP and external API calls (e.g., weather data) in templates/tools-basic.ts.
Sanitization: The skill provides a concrete example of a sanitizePrompt function in references/production-patterns.md to help developers mitigate injection risks.
[EXTERNAL_DOWNLOADS]: The skill documents fetching dependencies and configuration guidelines from well-known technology organizations.
Evidence: References to official documentation and repositories at ai-sdk.dev, vercel.com, github.com/vercel/ai, and github.com/cloudflare/ai. These are recognized as trusted organizations and well-known services and are documented neutrally as intended educational resources.
[COMMAND_EXECUTION]: Includes scripts and instructions for using package managers and tool generation CLIs during development.
Evidence: scripts/check-versions.sh utilizes standard npm commands to verify package versions. SKILL.md provides instructions for using npx mcp-to-ai-sdk to generate static tool definitions as a security best practice to avoid dynamic tool execution risks.

ai-sdk-core