The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project files and external Figma metadata, creating an inherent attack surface for indirect prompt injection if those files contain malicious instructions.
Ingestion points: scripts/resource-validator.py and scripts/responsive-validation.py read local codebase files (.tsx, .jsx, etc.) via file-system operations, and SKILL.md instructs the agent to fetch external data via Figma MCP servers (mcp0_get_design_context).
Boundary markers: Absent. The instructions do not supply prompt boundaries or instructions to ignore embedded natural language directives within the codebase or Figma specifications.
Capability inventory: File writing (open(..., 'w')) is performed in scripts/extract-and-implement.py and scripts/extract-component.py. Network access (requests.head) is implemented in scripts/resource-validator.py to check link integrity.
Sanitization: Absent. Content is processed via regex matching without filtering or context sanitization.
[DYNAMIC_EXECUTION]: The Python scripts (scripts/extract-and-implement.py and scripts/extract-component.py) perform script generation by assembling React/Tailwind component strings based on extracted design dimensions and writing them directly to the filesystem. This is a low-risk structural pattern tailored to the primary code-generation purpose of the skill.

pixel-perfect-ui