The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the AI agent to browse and analyze external inspiration websites such as Dribbble, Behance, and Awwwards using browser automation and multimodal AI tools. This creates a surface for indirect prompt injection where malicious instructions embedded in the visual or textual content of these third-party sites could attempt to influence the agent's behavior.
Ingestion points: references/workflows.md (Workflow 1 and Workflow 2) details the process of browsing external URLs and capturing screenshots for analysis.
Boundary markers: Absent. The skill provides no instructions to treat external data as untrusted or to ignore potential commands found within analyzed screenshots.
Capability inventory: The skill utilizes chrome-devtools for browser automation, bash (referenced for gallery-dl in references/design-resources.md), and media-processing (FFmpeg, ImageMagick) for file operations.
Sanitization: Absent. There are no defined verification or sanitization steps for content extracted from external platforms.
[COMMAND_EXECUTION]: The skill references the use of gallery-dl via bash in references/design-resources.md for downloading reference images. This indicates an expectation of shell command execution to process data derived from external, non-whitelisted sources.

aesthetic