The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided markdown and renders it into HTML templates using unescaped interpolation syntax, creating a surface for malicious content injection.\n
Ingestion points: Untrusted data enters via the markdown parameter defined in SKILL.md.\n
Boundary markers: Templates such as references/templates/explainer.html.tmpl and references/templates/incident.html.tmpl do not utilize delimiters or warnings to prevent the execution of instructions or scripts embedded in the data.\n
Capability inventory: The skill performs file system writes (output_path) and generates HTML that can reference external resources.\n
Sanitization: The templates use triple-brace interpolation (e.g., {{{body_html}}}), which renders HTML content without escaping, allowing potentially malicious scripts or styles to be included in the output reports.\n- [EXTERNAL_DOWNLOADS]: The skill references fonts and CSS resources from well-known content delivery networks to support Korean language rendering and design systems.\n
Evidence: Fetches fonts from Google Fonts (fonts.googleapis.com) and jsDelivr (cdn.jsdelivr.net).\n
Evidence: Optionally loads the Tailwind Play CDN to apply brand-specific styling when the design_system parameter is used.

doc-html-report