Control

This package appears to be a legitimate desktop automation skill that implements the features described, but there are a few red flags to consider before installing and running it: - Metadata mismatch: the registry owner/slug differ from the _meta.json inside the package. Confirm the author's identity and source before trusting the package. - Manual dependency installation: the skill expects pyautogui, opencv-python, pygetwindow, pyperclip, etc. Install these in a contained environment (virtualenv or VM) and inspect for platform-specific permission prompts (accessibility/access to control input on macOS, for example). - Powerful local capabilities: the skill can move your mouse, type into any active window, take screenshots, and read/modify clipboard contents. Only run it when you trust the code and on a non-sensitive session (or in a VM) and enable failsafe/require_approval. - Autonomy & LLM integration: the ai_agent component can execute multi-step tasks autonomously. Avoid providing an LLM client or credentials until you've reviewed any code paths that might send screenshots or other data to external services. - Safety tips: enable failsafe (DesktopController(failsafe=True)), prefer require_approval=True during testing, run demos step-by-step rather than 'run all', and test inside a disposable VM. If you plan to allow autonomous execution, review the full ai_agent.py to ensure it does not attempt to auto-detect or transmit credentials/data. If you want higher confidence: ask the publisher to reconcile the metadata, provide a provenance/source URL (repo or homepage), or provide a signed release. If you provide the truncated parts of ai_agent.py and __init__.py for review, I can re-evaluate any hidden behaviors.