The Agent Skills Directory

Indirect Prompt Injection: The skill ingests data extracted from external websites and uses it to generate scripts and creative direction. This creates a surface where content from the processed websites could potentially influence the agent's instructions.
Ingestion points: Data extracted from target URLs into the capture/ subfolder (specifically visible-text.txt and asset-descriptions.md) in references/step-1-capture.md.
Boundary markers: The skill does not currently specify the use of delimiters or clear separation between captured data and the agent's operational instructions.
Capability inventory: The skill environment allows for shell command execution (npx), file system writes, and network access.
Sanitization: There are no instructions for sanitizing or escaping the content retrieved from external websites before it is interpolated into scripts or code.
Dynamic Code Generation: The workflow involves generating HTML and JavaScript compositions in references/step-6-build.md based on the captured website data. These generated files are then executed in a browser-based environment for previewing and rendering. This process of creating and running code based on external inputs is a security consideration.
Command Execution: The skill utilizes shell commands via npx to perform core tasks like website capture, audio processing, and project validation. These operations are essential for the skill's purpose but involve the execution of external tools within the project environment.
External Downloads: The skill references external libraries from well-known content delivery networks (e.g., jsDelivr) for animation and rendering capabilities in references/techniques.md. While these are common development practices, they involve loading external scripts at runtime.

website-to-hyperframes