The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted design data from external Figma URLs using MCP tools. This data is used to generate production code and download assets, creating a surface where malicious instructions embedded in a design could influence agent behavior.
Ingestion points: External data fetched via get_design_context, get_metadata, and get_variable_defs from Figma design files.
Boundary markers: The skill does not define specific delimiters to isolate design data from instructions, although it provides a detailed multi-step workflow.
Capability inventory: File system writes for code and assets, network access via MCP tools, and font retrieval.
Sanitization: The workflow instructs the agent to manually re-express design logic into target components rather than direct verbatim copying.
[EXTERNAL_DOWNLOADS]: The skill downloads font files and CSS configuration from the Google Fonts API. These are standard operations for UI implementation targeting a well-known service.
[COMMAND_EXECUTION]: The skill executes a bundled script, scripts/verify-geometry.js, to perform pixel-level verification of the implemented layout. The script runs within the rendered page's environment to compare DOM dimensions against the design metadata.

figma-to-code