The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted design metadata and assets (component names, descriptions, styles, and SVG content) from the Figma API and incorporates them into generated React components, CSS, and TypeScript files. A compromised or malicious Figma file could theoretically influence the generated code or styles.
Ingestion points: Data is fetched via api.getFile, api.getImages, and fetch from the Figma REST API as shown in references/asset-export.md and references/component-generation.md.
Boundary markers: No explicit boundary markers or specific 'ignore instructions' delimiters are used when interpolating external design data into code templates.
Capability inventory: The skill uses fs.writeFile across several files to generate executable .tsx and .ts code.
Sanitization: While the skill performs naming normalization (e.g., normalizeFileName), it does not sanitize the contents of exported SVGs or component descriptions before embedding them in code strings.
[EXTERNAL_DOWNLOADS]: The skill downloads SVG files and design tokens from api.figma.com. Figma is a well-known and reputable service for design asset management, and these downloads are essential to the skill's primary purpose.
[COMMAND_EXECUTION]: The CI automation documentation in references/ci-automation.md describes the use of shell commands within GitHub Actions, such as npm install and npm run sync:design-tokens, to automate the synchronization process. These are standard developer practices.

figma-developer