The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on shell commands (ls, find, du, mkdir, mv) to analyze and reorganize the filesystem. In the instructions provided, placeholders like [target_directory] are sometimes shown without surrounding quotes (e.g., ls -la [target_directory]), which could lead to command injection if a directory name contains shell metacharacters.
[PROMPT_INJECTION]: As an indirect prompt injection surface, the skill processes untrusted input in the form of file names and directory structures. A maliciously named file (e.g., one containing shell commands or instructions intended to deceive the LLM) could potentially influence the agent's behavior during the analysis or organization phase.
Ingestion points: Reads directory listings, file names, and file type metadata via ls and find commands.
Boundary markers: None provided in the prompt to separate file metadata from agent instructions.
Capability inventory: Includes file-write (mkdir), file-move (mv), and file-deletion (via recommended removal of duplicates) capabilities.
Sanitization: No explicit sanitization or escaping of file names or paths is described in the instructions beyond a general recommendation to handle filename conflicts.

file-organizer