Dangerous exec
- Finding
- Shell command execution detected (child_process).
Security checks across static analysis, malware telemetry, and agentic risk
This appears to be a legitimate Linux desktop-automation skill, but it needs review because it can view and control the desktop and clipboard, relies on not-yet-validated gating, disables a built-in automation failsafe, and auto-installs runtime dependencies.
Treat this as granting an agent hands-on access to your Linux desktop. Prefer a VM or dedicated Linux/X11 account, review the source and dependencies, expect npm/pip runtime setup, keep sensitive windows and clipboard contents out of view, and verify an independent stop mechanism before using it on important accounts or production work.
VirusTotal findings are pending for this skill version.
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
If the agent targets the wrong window or acts without adequate review, it could click, type, launch apps, or use clipboard/screen data in unintended places.
The skill exposes broad desktop-control tools, and the README says a key containment check was not end-to-end validated on real Linux applications.
"standalone MCP server for screenshots, mouse, keyboard, app launch, window/display mapping, and clipboard" ... "Not yet validated in this session: ... foreground-window enforcement against real Linux apps"
Use only in a trusted, isolated Linux/X11 session; keep action confirmations enabled where possible; validate window-gating behavior before using it on sensitive accounts or production work.
A mistaken or looping automation sequence could interact with the desktop quickly without the standard PyAutoGUI mouse-corner stop mechanism.
The helper disables PyAutoGUI's built-in corner failsafe and removes action pauses, reducing the user's emergency-stop margin during automated mouse/keyboard activity.
pyautogui.FAILSAFE = False pyautogui.PAUSE = 0
Run in a VM, throwaway desktop session, or dedicated user account, and ensure you have an independent way to stop the process before granting it sensitive desktop access.
Installing or running the skill may download and install Python packages into the bundled runtime environment.
The runtime bootstrap is disclosed and aligned with the skill's purpose, but first-run dependency installation means the user is trusting the package sources used at runtime.
The first real run will automatically create `.runtime/venv` and install the public Python dependencies.
Review the dependency files, use trusted package indexes, and consider pinning/auditing dependencies before use in sensitive environments.
The skill can spawn local programs as part of building/running its desktop-control runtime.
The static scan shows local process execution. That is expected for this build/runtime bridge, but it is still a powerful local capability.
const child = execFile(
Only run the bundled project from a source you trust and avoid granting it access to sensitive desktop sessions until you have reviewed the code and dependencies.
Whatever is visible on screen or present in the clipboard may be processed by the local agent session using this MCP server.
Screenshots and clipboard contents are sensitive local data that can be exposed to the connected MCP client/agent. The artifacts frame this as trusted-local use, which is purpose-aligned but important for users to understand.
standalone MCP server for screenshots, mouse, keyboard, app launch, window/display mapping, and clipboard
Use the skill only with trusted local agents, and clear or hide sensitive windows and clipboard contents before running desktop automation.