The Agent Skills Directory

[DYNAMIC_EXECUTION]: The documentation and example implementations (specifically in references/tool-usage.md and references/library-guide.md) demonstrate the use of the eval() function for implementing calculator-style tools. While these examples include comments suggesting the use of a safe evaluation library for production, the direct inclusion of eval() in the logic represents a risk for dynamic code execution if the patterns are implemented as provided without further modification.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from multiple sources, including free-text user input, file uploads, and AI-generated streaming content. This architecture creates an attack surface for indirect prompt injection.
Ingestion points: textarea input fields, image and file upload components, and API-driven streaming text blocks found in SKILL.md and example files like examples/basic-chat.tsx.
Boundary markers: The skill suggests using system prompts (available in assets/system-prompts.json) to define agent behavior and outlines strategies for communicating AI limitations and context status to the user.
Capability inventory: The skill utilizes network access to AI providers (via Vercel AI SDK), local storage through IndexedDB for persistence, and various tool/function calling capabilities.
Sanitization: The skill follows security best practices by explicitly recommending and providing code examples for DOMPurify to sanitize AI-generated markdown and HTML before rendering.
[EXTERNAL_DOWNLOADS]: The skill references and requires several external dependencies from well-known technology companies and trusted organizations, such as Vercel (AI SDK and Streamdown) and the DOMPurify project. These resources are standard for the intended development use case.

building-ai-chat