The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process external image files and inspect their metadata using magick identify. This creates a vulnerability surface for indirect prompt injection, as malicious instructions hidden in image metadata (like EXIF or IPTC tags) could be read by the agent and influence subsequent actions.
Ingestion points: Image metadata extracted via magick identify in SKILL.md and references/recipes.md.
Boundary markers: None specified in the provided command patterns or instructions.
Capability inventory: The skill utilizes the magick CLI which has capabilities for file system access (read/write) and execution of external delegates (as noted in references/security.md).
Sanitization: There are no instructions provided to sanitize or filter metadata content before the agent processes the output of identify.

imagemagick