The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill makes outbound calls to the official Figma REST API (api.figma.com) to fetch design documents, styles, and images. This is the intended behavior for a Figma analysis tool.
[COMMAND_EXECUTION]: The skill executes several shell commands during its lifecycle:
Phase 0: Uses pnpm install and pnpm build to prepare the environment and build the @understand-anything/core library.
Phase 1 & 3: Executes Node.js scripts (figma-scan.mjs, figma-merge.mjs) to process design data.
Phase 4: Uses find and rm to clean up intermediate JSON files in the .ua/intermediate directory.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8):
Ingestion points: Design data (layer names, component names, and text content) is fetched from the Figma REST API and passed to an LLM subagent for analysis.
Boundary markers: The instructions for batching design nodes into subagents do not specify the use of clear delimiters or instructions to ignore embedded commands within the design content.
Capability inventory: The skill environment has access to shell commands (node, pnpm, find, rm) and file system write access within the project directory.
Sanitization: There is no evidence of sanitization or filtering of the content retrieved from the Figma API before it is processed by the LLM subagent.
[CREDENTIALS_UNSAFE]: The skill requires a FIGMA_TOKEN. The instructions correctly advise the user to provide this via an environment variable and explicitly warn against writing the token to logs or files. The token is passed to the scanning script as an environment variable.

understand-figma