The Agent Skills Directory

[SAFE]: The skill is entirely instructional, providing best practices for website information architecture without executing code or accessing sensitive resources.
[DATA_EXPOSURE]: The skill instructions suggest reading local context files like .agents/product-marketing.md. This is a standard design pattern for agents to gather project-specific knowledge and does not constitute a security risk, as it involves local workspace data intended for the agent's use.
[INDIRECT_PROMPT_INJECTION]: The skill has an ingestion surface for untrusted data via the aforementioned marketing context files. However, because the skill defines a text-generation role without access to dangerous system tools (e.g., shell access, network exfiltration), the risk is negligible.

site-architecture