The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by scanning and processing files within user-provided directories. This behavior creates an attack surface where maliciously crafted filenames or metadata within image files could be used to influence the agent's behavior if the agent later interprets these strings as instructions.
Ingestion points: The script scripts/resize_images.py reads a list of files from a directory provided as a command-line argument and opens each file using the Pillow library.
Boundary markers: The instructions do not define clear boundaries or delimiters to separate data (filenames/metadata) from the agent's operating instructions.
Capability inventory: The skill has the ability to execute shell commands (Bash), read files, and write resized images to the disk.
Sanitization: The script performs structural validation by checking file extensions and using a standard image processing library, but it does not sanitize text-based metadata or filenames that may be displayed back to the agent.

resize-images