pipx-desktop-agent

• Risk level: Low
• Analyzed: Feb 24, 2026, 5:13 AM
• Findings: 0
• Version: 1.3.2

Overview

The package and its runtime instructions are coherent with a desktop automation tool: code, CLI docs, and dependencies align with the described mouse/keyboard/screen capabilities, but it grants powerful local control (screenshots, typing, subprocesses) so only install from a trusted source and on a machine you permit those actions on.

This package appears to be what it claims: a full-featured desktop automation CLI using PyAutoGUI and OCR. Before installing or enabling it, consider: - Trust: only install if you trust the publisher/source. The registry metadata omitted an install spec but the repo contains install instructions (pipx/pip) and code—treat it like a third‑party package. - Powerful local access: it can move your mouse, type keys, take screenshots (saved to disk), enumerate and focus windows, and present dialogs that can capture user input (including passwords). These are expected for automation but can expose sensitive data. - Subprocess risk: the app.open implementation uses subprocess calls and on Windows constructs a shell command with shell=True; avoid passing untrusted strings to it (risk of shell injection). - Dependencies: OCR/vision features require heavy native/third-party dependencies (easyocr, possibly torch, and platform tools like tesseract or wmctrl/xdotool). Installing may pull large packages and require extra system packages. - Mitigations: run in a controlled environment (VM/container) if you are unsure, review the code (already included) yourself, and restrict use to tasks you explicitly authorize. If you need the functionality but worry about sensitivity, consider running the tool locally under your control (not giving remote agents blanket access) or creating policies that require user confirmation for sensitive commands (screenshots/password prompts).