The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and fetches visualization libraries such as Chart.js, D3.js, and Mermaid from established and well-known CDN providers including cdnjs.cloudflare.com, esm.sh, cdn.jsdelivr.net, and unpkg.com.
[REMOTE_CODE_EXECUTION]: The skill is designed to generate and execute HTML, CSS, and JavaScript inside a sandboxed iframe. The implementation guidelines provided in the skill documentation specify the use of sandbox="allow-scripts" and a restrictive Content Security Policy (default-src 'none') to ensure the generated code remains isolated from the main application environment and cannot perform unauthorized network operations.
[INDIRECT_PROMPT_INJECTION]: The system implements a sendPrompt bridge allowing scripts within the visualization to communicate with the chat agent, which creates a potential surface for indirect prompt injection.
Ingestion points: User queries requesting visualizations or diagrams (processed in SKILL.md).
Boundary markers: Visual content is isolated within visualizer code blocks.
Capability inventory: Provides sendPrompt (sending messages back to the chat) and openLink (external link navigation) capabilities.
Sanitization: Relies on the agent's internal safety guardrails for generation and the host's iframe sandbox for execution; no specific sanitization logic is provided within the skill instructions.

visualise