The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection surface identified. The skill ingests untrusted content from external Figma files and URLs using tools such as get_design_context and get_libraries. Malicious instructions embedded within Figma layer names or metadata could influence agent behavior during processing.
Ingestion points: External Figma file keys, node IDs, and design context data retrieved via MCP tools.
Boundary markers: No explicit delimiters or instructions are provided to ignore embedded commands in the design data.
Capability inventory: Executes file writes to design-state.md and generates HTML code for prototypes.
Sanitization: There is no mention of sanitizing or validating the content pulled from Figma before it is processed.
[COMMAND_EXECUTION]: Dynamic code generation surface. The skill generates a self-contained, clickable HTML prototype at runtime when Figma tools are unavailable. This process involves assembling code from project tokens and content strings, which is a form of dynamic execution.
[COMMAND_EXECUTION]: The skill invokes various Figma-specific MCP tools (e.g., use_figma, create_new_file) to perform design operations. This interaction with an external platform through shell-like commands is a core capability of the skill.

figma-bridge