The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted natural language prompts via the generate_ui tool to create UI specifications. This capability includes a CustomHTML component that can execute JavaScript, presenting a potential surface for indirect injection. However, the risk is inherent to the primary purpose of the skill and is well-mitigated through strict JSON enforcement, HTML sanitization, and execution within a sandboxed iframe.
Ingestion points: generate_ui tool documented in SKILL.md.
Boundary markers: System prompts used in the generation pipeline enforce strict JSON output formats.
Capability inventory: JavaScript execution via CustomHTML in daub-render.js.
Sanitization: Implementation includes an allowlist-based sanitizeHtml function and asset validation via isSafeUrl, and previews are rendered in a sandboxed iframe.
[EXTERNAL_DOWNLOADS]: Retrieves design assets, icons, and fonts from established and reputable services including JSDelivr, Unpkg, and Google Fonts. These resources are well-known technology services and are documented neutrally.

daub-ui