The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided images, which creates an ingestion surface for potentially untrusted data. * Ingestion points: Images are selected via the asset_add_file tool or provided via local paths and Creative Cloud URNs in non-UI fallback scenarios (SKILL.md). * Boundary markers: The instructions do not define specific delimiters or security warnings to prevent the agent from acting on instructions that might be embedded in image metadata. * Capability inventory: The skill utilizes various image manipulation tools (image_auto_straighten, image_apply_adjustments, image_fill_area) and shell commands (curl) for file management. * Sanitization: There are no explicit steps described for sanitizing image metadata or validating the source of external URLs.
[COMMAND_EXECUTION]: The skill provides instructions to use curl for downloading processed images into the local workspace in environments without a graphical user interface (Steps 2c and 8). These operations target Adobe's official storage infrastructure and are part of the intended asset management workflow.

adobe-retouch-portraits