The Agent Skills Directory

Command Execution of Workspace Tools: The skill executes several Node.js scripts located in /workspace/tools/ for rendering HTML, linting brand assets, and managing project files. These tools are pre-installed in the environment and are central to the skill's operation.
External Asset Downloads: The skill includes functionality via fetch-asset.mjs to download image assets from external URLs provided at runtime. This introduces a potential vector for retrieving untrusted content into the workspace.
Indirect Prompt Injection Surface: The skill processes external data (copy and images) that acts as a potential injection vector.
Ingestion points: Untrusted content enters the agent's context through user-provided copy and external image URLs.
Boundary markers: The instructions do not define specific delimiters or instructions for the agent to separate untrusted content from the system prompts.
Capability inventory: The skill has access to shell command execution (via node), file system write operations, and asset export capabilities.
Sanitization: The skill lacks explicit sanitization or escaping requirements for external text before it is interpolated into HTML templates.
Dynamic Content Generation: The skill programmatically generates HTML which is then rendered by a headless browser tool. This involves the dynamic generation and execution of code based on project requirements, which is a standard but noteworthy pattern in design-oriented skills.

stills