The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external URLs via tools like extract-page-data or clone-page. This data (theme.json and screenshots) is then used to populate the generated DESIGN.md and globals.css. Maliciously crafted metadata or CSS variable names on a target website could attempt to influence the agent's behavior during the generation phase.
Ingestion points: theme-data/theme.json and theme-data/meta.json (scripts/generate.mjs).
Boundary markers: The skill instructions emphasize that "screenshots are the final truth" and explicitly warn that theme.json may contain machine-extracted noise, encouraging cross-referencing.
Capability inventory: The skill writes files to the local filesystem (DESIGN.md, globals.css) and copies image files.
Sanitization: The generate.mjs script performs basic parsing (JSON.parse) and color validation (rgbToHex) but does not rigorously sanitize string values (like font names or variable names) before interpolating them into Markdown and CSS files.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands using external tools (node <extract-page-data>/scripts/extract.mjs). While these are intended for data collection, they involve passing user-provided URLs directly to shell scripts.

generate-theme