The Agent Skills Directory

[COMMAND_EXECUTION]: The figma_text.py script executes a local Node.js script (figma_cmd.mjs) using subprocess.check_output. This is a functional requirement to bridge with the Figma WebSocket API. The execution uses a list-based argument passing method, which avoids shell injection vulnerabilities within the script itself.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text from Figma design nodes. While this data could contain instructions intended to influence the agent, the skill implements a mandatory user review step where all proposed text changes are presented in a diff table for approval before any modifications are applied back to Figma. Ingestion occurs in scripts/figma_text.py via the extract command, and capabilities include design modification via the apply command.
[DATA_EXPOSURE]: Extracted design text is temporarily stored in /tmp/figma_texts.json. This is standard behavior for passing data between tool steps in a local environment and does not involve data exfiltration to external domains.

content-review