The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute Node.js scripts (evaluate.mjs, typography-evaluate.mjs) locally to perform deterministic checks on design data. It also utilizes the use_figma tool to run code within the Figma plugin context.
[DYNAMIC_EXECUTION]: The skill uses a template-based approach for Figma data extraction where the agent substitutes constants into a JavaScript file (scripts/extract.figma.js) before execution. This dynamic script generation is constrained by instructions that forbid rewriting the core logic, but it represents a runtime code assembly pattern.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted content from Figma designs (such as node names and text values). This data is used by the agent during semantic evaluation steps, creating a potential vector for malicious content to influence the LLM's output.
Ingestion points: Figma node names, text characters, and layout metadata retrieved from the Figma API and stored in a temporary JSON file.
Boundary markers: The agent is directed to use a specific rubric derived from the design system schema and to rely on predefined scripts rather than generating evaluation logic from scratch.
Capability inventory: Execution of JavaScript in the Figma environment (use_figma), local Node.js command execution, and local file system writes to the /tmp directory.
Sanitization: The skill uses deterministic scripts to extract structured data from Figma, providing a layer of separation between content and logic, though the text content itself is not sanitized before being analyzed by the LLM.

figma-token-review