The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves design data and textual content from Figma via mcp__figma__get_design_context and mcp__figma__get_metadata. As Figma content is user-controlled, this represents a surface for indirect prompt injection. \n
Ingestion points: Design tokens and strings are fetched from external Figma URLs provided as input. \n
Boundary markers: The skill does not define specific boundaries or instructions to isolate external design data from the agent's core instructions. \n
Capability inventory: The agent has access to Bash for searching the local filesystem and Playwright for browser interaction. \n
Sanitization: There is no explicit sanitization process for the data retrieved from the design specification before it is used for comparison. \n- [COMMAND_EXECUTION]: The skill includes Bash recipes using grep to scan local source files (features/, components/, src/) for development-related strings such as 'TODO' or 'placeholder'. \n- [DATA_EXFILTRATION]: The skill uses Playwright to capture screenshots of the local development environment and surfaces these images, along with Figma design data, within the interaction transcript for visual evaluation. This allows implementation details to be processed by the LLM provider for the purpose of visual diffing.

figma-verify