OpenClaw with Apple
Warn
Audited by Gen Agent Trust Hub on Mar 18, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The
SKILL.mdfile contains 'Iron Rules' (铁律) that explicitly instruct the AI agent to bypass standard conversational confirmations and execute commands immediately upon detecting task-related keywords. This overrides typical safety and helpfulness behaviors by forcing immediate execution and prohibiting clarification questions. - [CREDENTIALS_UNSAFE]: The skill instructs the user to provide their primary Apple ID email and main password to the AI. The scripts then store these credentials in environment variables (
ICLOUD_PASSWORD) to facilitate non-interactive login via thepyicloudlibrary. Handling primary account passwords in plain-text environment variables is a high-risk security practice. - [COMMAND_EXECUTION]: Multiple scripts, including
setup_tasks_cron.py,icloud_tool.py, andstatus_wall.py, use thesubprocessmodule to dynamically execute other Python scripts and system utilities likelaunchctlfor task scheduling and status monitoring. - [EXTERNAL_DOWNLOADS]: The skill requires the installation of external dependencies (
pyicloud,caldav,icalendar) from the official Python Package Index (PyPI). It also references official Appleicloud.comURLs for iOS Shortcut imports, which are considered well-known and safe sources.
Audit Metadata