The Agent Skills Directory

[COMMAND_EXECUTION]: Potential shell command injection vulnerability. The skill constructs commands like exiftool -Artist="<artist>" where <artist> is user-provided input. If the input contains shell metacharacters (e.g., backticks, dollar signs, or semicolons), it could lead to arbitrary command execution.
[COMMAND_EXECUTION]: Suggests the use of sudo apt install for dependency management. While this is a standard installation procedure, the use of sudo within an agent's execution context should be monitored for unintended privilege escalation.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting untrusted data (metadata tag values) and writing them into files.
Ingestion points: Metadata key-value pairs provided by the user (e.g., Artist, Copyright, Description) as described in the 'Inputs to gather' section.
Boundary markers: Absent. User input is directly placed into command-line arguments.
Capability inventory: The skill uses Bash with access to exiftool, ls, find, cp, and rm.
Sanitization: Absent. No instructions are provided to the agent to escape, validate, or sanitize the user-provided strings before processing.

set-metadata