The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied file paths and incorporates user-defined parameters (format, size, quality) directly into command-line arguments. While intended for image processing, this creates a surface for processing unexpected data if an attacker provides paths to sensitive system files.
Ingestion points: The skill accepts a list of image paths from the user context (SKILL.md).
Boundary markers: None present. There are no explicit instructions to ignore embedded content or validate that the provided paths are actually images before calling the script.
Capability inventory: The skill executes a Python script (scripts/compress_images.py) which uses the Pillow library to read and write files to the local file system.
Sanitization: The script uses Path(image_path_str).expanduser() and argparse for argument handling, but does not strictly validate the MIME type or content of the files before attempting to open them with Image.open().
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute specific Python commands via the terminal. This is a standard functional pattern for this type of utility but involves direct shell interaction.
[SAFE]: The skill relies on well-known, standard libraries (pillow, python-dotenv) from official package registries. The code logic for image resizing and compression is transparent and matches the described purpose.

image-compressor