The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to write a file to a temporary directory and immediately execute platform-specific opening commands (xdg-open, open, or start). This results in the automatic execution of the generated file in the user's default browser without prior user review of the source code.
[REMOTE_CODE_EXECUTION]: The instructions require the generation of HTML that includes external JavaScript from well-known content delivery networks (Tailwind and Mermaid). Crucially, it mandates configuring Mermaid with securityLevel: "loose". This configuration explicitly disables the Mermaid renderer's internal sandbox, allowing the rendering of potentially malicious HTML or scripts if they are included in diagram labels.
[EXTERNAL_DOWNLOADS]: The skill fetches libraries from cdn.tailwindcss.com, Google Fonts, and cdn.jsdelivr.net. These are recognized well-known services used for UI rendering and diagram support.
[PROMPT_INJECTION]: The skill facilitates an indirect injection attack surface. It ingests data from the conversation history to populate the HTML file but lacks explicit instructions for sanitizing or escaping this content. If an attacker provides malicious HTML or script tags in the chat, the agent will embed them into a file that is then automatically executed with a lowered security configuration.

html-output