The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests untrusted data from Figma node names, design hierarchies, and user comments (as seen in references/intake-figma-sketch.md). While it includes a boundary marker in SKILL.md (Hard Rule 7) instructing the agent to treat this as data, it possesses significant capabilities including shell command execution (curl, jq, python3) and the ability to write back to Figma via mcp__figma__post_comment. No explicit sanitization or strict schema validation is defined for the ingested design data.\n- [DYNAMIC_EXECUTION]: In references/hig-general-checklist.md, the skill defines a Python script to be executed via a heredoc for calculating color contrast ratios. This constitutes the dynamic execution of locally provided code to perform design verification.\n- [COMMAND_EXECUTION]: The skill instructs the agent to use various CLI tools (curl, jq, grep, python3) to fetch guidelines from external servers and process data stored in local MCP cache files.\n- [EXTERNAL_DOWNLOADS]: The skill fetches official documentation from Apple's developer domain (developer.apple.com) and utilizes the mcp-figma package via npx to interface with Figma. These references to well-known services are documented as part of the skill's operational flow.

hig-design-review