The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and fetches standard, well-known software packages and documentation from the TanStack organization, as well as official SDKs and models from established AI providers including OpenAI, Anthropic, and Google.
[PROMPT_INJECTION]: The skill facilitates the construction of AI agents that ingest untrusted data from users and external tool outputs, establishing an inherent surface for indirect prompt injection attacks.
Ingestion points: External data enters the agent context via the messages array in server-side chat() functions and through user input processed by the useChat hook in framework-specific packages.
Boundary markers: The skill utilizes the structured UIMessage format with discrete parts (text, image, tool-call), which provides a logical separation of content types but does not explicitly enforce security-specific delimiters to mitigate instruction injection within those parts.
Capability inventory: Agents implemented using these patterns possess significant capabilities, including performing arbitrary network requests via connection adapters and executing potentially sensitive operations through server and client tools.
Sanitization: No specific content sanitization or automated filtering of incoming message parts is mandated by the architectural patterns described in the skill documentation.

ai-core