The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/disk_optimizer.py utilizes subprocess.run to execute system utilities du (for accurate disk usage calculation) and rsync (for transactional file migration). These executions are implemented using argument lists rather than shell strings, effectively preventing command injection vulnerabilities. The commands are scoped to the skill's primary purpose of disk management.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect injection as it scans and processes local file system metadata (file and directory names).
Ingestion points: The inventory command in scripts/disk_optimizer.py reads directory structures and file names to identify cleanup candidates.
Boundary markers: The skill uses structured JSON output for inventory and planning results, and relies on an explicit multi-step workflow (inventory -> plan -> execute).
Capability inventory: The skill has the capability to move files to the Trash and delete specifically marked .cleanup items from the Trash.
Sanitization: The script includes a hardcoded PROTECTED_PARTS list that excludes sensitive directories (e.g., .git, Keychains, Mail, Messages) and applies dual-path normalization (resolved and lexical) to ensure file operations do not escape intended boundaries.
[SAFE]: The skill implements a 'transactional' migration model (Copy -> Verify -> Move to Trash -> Link), ensuring that source data is only removed after a successful integrity check. It also features a specific protection layer for application-managed recording workspaces (e.g., Screen Studio), preventing accidental deletion of project-critical metadata.

lov-clean-mac