The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external image files (PNG, JPEG, WebP, etc.) to extract metadata, color palettes, and structural information. This creates an attack surface where malicious instructions could be embedded within image metadata (EXIF, ICC profiles) or hidden visually within the image to influence the agent's behavior.
Ingestion points: analyze.py and boxes.py load user-provided image files using Image.open().
Boundary markers: The instructions do not define specific delimiters or warnings to prevent the agent from following instructions found inside images.
Capability inventory: The skill executes local Python scripts via subprocess to perform file analysis and color extraction.
Sanitization: There is no sanitization of extracted metadata strings before they are returned to the agent.
[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands to set up a Python virtual environment and install external dependencies (pillow, numpy, and scipy). While these are standard libraries, the execution of shell commands and package managers represents a common vector for environment manipulation if the instructions were to be modified.

image-analysis