The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements code generation templates (e.g., generateReactComponent, generateJSX) that are susceptible to indirect injection because they ingest untrusted data from an external source and interpolate it into executable code files.
Ingestion points: Design data is fetched from the Figma REST API in getComponentCode, getAllComponents, and extractTypography within SKILL.md.
Boundary markers: The code generation logic lacks boundary markers or explicit instructions to ignore potentially malicious content embedded within Figma layers or text nodes.
Capability inventory: The figmaToCode function uses fs.writeFile to write the generated React, CSS, and SVG files to the local filesystem.
Sanitization: The implementation lacks sanitization; external strings such as node.name and node.characters are directly used as component names and JSX children, which could lead to code injection if a Figma file is maliciously crafted.
[COMMAND_EXECUTION]: The design-to-code pipeline documented in SKILL.md utilizes the fs.writeFile method to perform persistent file system writes. This capability allows the skill to create or overwrite files on the local machine based on the results of external API calls.

figma