The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local shell commands (ls, find, du, md5, sed) to inspect the file system, calculate hashes for duplicate detection, and calculate disk usage. These commands are restricted to the local environment and are used to provide the user with a current state analysis and organization plan.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests untrusted data from the local file system.
Ingestion points: SKILL.md (Step 2 and 4) instructs the agent to read directory listings, file names, and the first 20 lines of files (head -20).
Boundary markers: The skill implements strict boundary markers by requiring a "Current state summary" and a "Proposed Organization Plan" that must be explicitly approved by the user before any file modifications or deletions occur.
Capability inventory: The skill has the capability to move, rename, and delete files based on the generated plan.
Sanitization: No technical sanitization of file contents is mentioned, but the human-in-the-loop approval requirement serves as the primary control mechanism.
[COMMAND_EXECUTION]: The instructions mandate strict safety rules for file operations, including logging all moves for potential undo, preserving modification dates, and handling filename conflicts gracefully, which mitigates risks associated with automated file system management.

file-organizer