The Agent Skills Directory

[COMMAND_EXECUTION]: The utility uses the subprocess module to execute system commands including ps for process monitoring, security for keychain management, and systemsetup for configuration. These are restricted to specific, hardcoded arguments related to the skill's stated purpose.
[PRIVILEGE_ESCALATION]: The script requests administrative privileges via sudo to modify the system timezone. It employs sudo -v to ensure the user is prompted for credentials by the operating system, and the functionality is explicitly disclosed to the user before execution.
[INDIRECT_PROMPT_INJECTION]: The tool processes local configuration data. 1. Ingestion points: Reads ~/.claude/settings.json and ~/.claude.json in scripts/claude_cleanup.py. 2. Boundary markers: Employs mandatory user CONFIRM input and strictly delimited key-based modifications. 3. Capability inventory: Includes file manipulation, process listing, and administrative system configuration. 4. Sanitization: Uses JSON schema validation and a PROTECTED whitelist to prevent tampering with sensitive agent assets.

claude-cleanup