The Agent Skills Directory

Potential Indirect Prompt Injection Surface: The skill is designed to process external content, including live website URLs and visual mockups, to generate code and prototypes. This functionality creates an entry point where external data containing instructions could potentially influence agent behavior.
Ingestion points: Processes live URLs through the $url-to-code workflow and captures product screens via the browser tool (SKILL.md).
Boundary markers: The instructions do not define specific delimiters or "ignore" directives to isolate external content from the core instructions.
Capability inventory: The skill interacts with browser tools and external deployment platforms (e.g., @Vercel, @Sites) as noted in the $share and browser choice sections (SKILL.md).
Sanitization: No explicit sanitization or filtering logic is mentioned for content retrieved from live URLs or analyzed visual sources.
Local Script Execution Pattern: The skill includes instructions to execute a "preflight script" located within the skill's file structure (../user-context/SKILL.md) when shell access is available. This pattern is typically used for initialization and environment verification but involves active command execution on the host environment.

index