The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data retrieved from Figma designs (via get_design_context and get_metadata). While this presents a theoretical attack surface where an adversary could embed instructions in Figma layer names or metadata, the skill focuses strictly on translating visual and structural properties into code. This risk is inherent to the skill's primary purpose of design-to-code translation and is mitigated by the structured tool outputs and specific instructions for code generation.
Ingestion points: Data returned from Figma MCP tools (get_design_context, get_metadata) based on user-provided Figma URLs.
Boundary markers: None explicitly defined for the tool output interpolation.
Capability inventory: The skill facilitates design data fetching and visual reference capturing; it does not introduce additional high-risk capabilities beyond standard agent file operations.
Sanitization: None explicitly mentioned for the Figma design data.
[EXTERNAL_DOWNLOADS]: The skill retrieves assets (images, SVGs) exclusively through the Figma MCP server's designated assets endpoint (often serving from localhost). This is a legitimate and controlled method for asset acquisition within the Figma MCP ecosystem.
[COMMAND_EXECUTION]: The instructions utilize standard MCP tool calls (get_design_context, get_screenshot, get_metadata) which are appropriate for interacting with the Figma API via a local or remote MCP server.

implement-design