The Agent Skills Directory

[METADATA_POISONING]: The skill metadata and license file identify the author as "AgriciDaniel", which conflicts with the system-identified author "flexksx". This inconsistency in authorship documentation is misleading.
[COMMAND_EXECUTION]: The skill relies on shell command execution for its primary functions, including image conversion and metadata manipulation using tools such as exiftool, cwebp, convert, and ffmpeg. These tools are invoked with parameters that may be derived from external sources.
[PRIVILEGE_ESCALATION]: The documentation explicitly instructs users to execute commands with elevated privileges (sudo apt install) to set up necessary system dependencies like libimage-exiftool-perl and webp.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external URLs, including HTML image attributes and file metadata, which is then used as input for further analysis and as variables in command execution.
Ingestion points: HTML <img> tags, alt text, and image metadata (IPTC/XMP) retrieved from external websites during audits.
Boundary markers: No specific boundary markers or instructions to treat external data as untrusted are present in the skill instructions.
Capability inventory: Access to shell commands (exiftool, cwebp, etc.) and file system interaction for image optimization and format conversion.
Sanitization: The provided instructions do not define any sanitization, escaping, or validation logic for the content fetched from external sources before it is processed.

seo-images