The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process product descriptions, category data, and user-provided lists to generate SEO-optimized content and HTML templates. This creates a surface where malicious instructions embedded in product data could influence the agent.
Ingestion points: Data retrieved via shop_read (get, list, category_list) in SKILL.md and references/API_REFERENCE.md, as well as external product lists provided by users.
Boundary markers: The instructions lack explicit delimitation (e.g., XML tags or clear delimiters) for untrusted content during prompt interpolation.
Capability inventory: The skill possesses significant write capabilities across the shop module, including shop_write for product/category updates and templates_write for modifying site code.
Sanitization: The references/COPY_GUIDE.md provides guidance on using semantic HTML and preserving system variables, which serves as a rudimentary form of output sanitization.
[COMMAND_EXECUTION]: The skill utilizes a custom tool call for telemetry and usage accounting.
Evidence: The instruction skills_tool(action="register_usage", skill_id="ucoz-shop-optimizer-skill") in SKILL.md.
Context: This is a vendor-specific usage tracking mechanism. While it involves a tool execution, it targets an internal platform capability for the vendor and does not execute arbitrary shell commands.
[SAFE]: The skill includes explicit security constraints that forbid the AI from requesting uAPI keys or using unauthenticated network paths, which demonstrates a high degree of security awareness in the skill design.

ucoz-shop-optimizer-skill