The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The scripts/compose.ts script interpolates content from inputs.json directly into the generated HTML without sanitization. The schema.ts documentation explicitly mentions that certain fields can include raw HTML. This creates an attack surface where malicious instructions or scripts embedded in processed data could be executed if the agent is directed to process untrusted external content.
Ingestion points: inputs.json (read by scripts/compose.ts).
Boundary markers: Absent. No explicit instructions are provided to the agent or the script to ignore or escape embedded instructions in the slide content.
Capability inventory: file-write (writing the output HTML file via node:fs/promises).
Sanitization: Absent; the mixed helper and various render functions in scripts/compose.ts interpolate content directly into HTML strings without escaping.
[EXTERNAL_DOWNLOADS]: Fetches typography assets from Google Fonts. The generated HTML and the scripts/compose.ts renderer reference fonts.googleapis.com and fonts.gstatic.com to load the "Inter Tight", "Playfair Display", and "JetBrains Mono" font families.

open-design-landing-deck