The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents standard package installation commands for necessary dependencies like micromark and the openai SDK, which is consistent with its primary purpose.\n- [EXTERNAL_DOWNLOADS]: The skill references established and well-known services, including OpenAI and Azure for AI integrations, and the esm.sh Content Delivery Network for loading JavaScript modules.\n- [DATA_EXFILTRATION]: No evidence of unauthorized data exfiltration was found. The examples demonstrate secure practices by using environment variables (e.g., import.meta.env) for handling sensitive API keys rather than hardcoding them.\n- [PROMPT_INJECTION]: The skill identifies the potential for indirect prompt injection via AI-generated content and explicitly instructs developers to sanitize output using libraries like DOMPurify before rendering it to the DOM.\n
Ingestion points: User input captured via onMessageEntered and responses received from external AI APIs.\n
Boundary markers: The instructions recommend clear separation of user and bot messages using stable identifiers.\n
Capability inventory: The skill provides examples of rendering HTML content through framework-specific properties such as [innerHTML], v-html, and dangerouslySetInnerHTML across all references.\n
Sanitization: Explicitly recommends using DOMPurify to sanitize HTML content before rendering to mitigate XSS and injection risks.

devextreme-chat