The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions and reference patterns facilitate the ingestion of untrusted data from Figma canvas nodes (such as names, text content, and descriptions) which are attacker-controlled.
Ingestion points: Metadata extraction and discovery scripts in references/component-patterns.md and references/variable-patterns.md read properties from Figma nodes via findAllWithCriteria, node.query, and figma.root.children.
Capability inventory: The use_figma tool enables arbitrary JavaScript execution within the Figma Plugin API context, and the upload_assets tool provides network POST capabilities to tool-generated URLs.
Sanitization: The provided reference patterns and rules do not include guidance for sanitizing or escaping the content read from Figma nodes before it is processed by the agent or used in subsequent code generation.
Boundary markers: There are no specific instructions regarding the use of boundary markers or 'ignore embedded instructions' warnings for data ingested from the Figma environment.

figma-use