The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and process data from external image files and folders.
Ingestion points: Commands such as import_folder and import_files (SKILL.md) bring external file metadata and content into the agent's processing context.
Boundary markers: Absent. The instructions do not direct the agent to treat imported data as untrusted or to use specific delimiters when processing file content.
Capability inventory: The skill allows for file-system writes (export_images), execution of the cull binary, and interactive curation through various MCP tools.
Sanitization: Absent. There is no mention of sanitizing or validating image metadata or content before it is processed by the agent.

cull