The Agent Skills Directory

[SAFE]: The skill implements a secure-by-design approach by requiring the agent to map all user inputs to a strict 'closed vocabulary' of symbolic values defined in Section 2.1. Any unrecognized values must trigger a clarification prompt rather than being processed, which serves as an effective safeguard against prompt injection and unexpected behavior.
[DATA_EXPOSURE]: The skill requests the file_write capability, which is used solely to generate a DESIGN.md specification and a brief-preview.html file within the working directory. No access to sensitive system paths (e.g., .ssh, .aws) or hardcoded credentials was identified.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied design briefs, creating a potential attack surface. However, this risk is mitigated by the following evidence chain: 1) Ingestion point is the brief input string in SKILL.md. 2) Boundary markers are established via the strict symbolic-to-token resolution table. 3) Capabilities are limited to local file writing. 4) Sanitization is performed through the mandatory mapping logic that rejects input not matching the predefined design system vocabulary.

design-brief