The Agent Skills Directory

[SAFE]: The skill functions as a structured pipeline orchestrator for design tasks, following a documented workflow from onboarding to delivery.
[DATA_EXPOSURE]: The skill reads the local codebase to extract design tokens (e.g., CSS variables, Tailwind configurations). This data is persisted in a local directory (.agents/design-system/) to provide a consistent design system across multiple agent sessions.
[EXTERNAL_DOWNLOADS]: Mentions the use of external services for asset generation (e.g., Gemini API, Meshy, Suno). These operations are restricted by an 'Ask First' policy, requiring explicit user consent before execution.
[COMMAND_EXECUTION]: Delegates tasks to sub-agents using platform-specific tools like Agent or spawn_agent. This is the core mechanism for its orchestration capabilities.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes external content like codebase files and user briefs. It mitigates this risk by using structured data schemas (DESIGN_INTENT_HANDOFF) for delegation and implementing a 'Warden' pre-check gate for major operations.

atelier