Back to skill

Security audit

Desktop Control

Security checks for vulnerabilities and agentic risk

Overview

This desktop automation skill is not deceptive, but it gives broad screen, clipboard, keyboard, mouse, exec, and optional administrator-level control with weak default safeguards.

Install only if you intentionally want an agent to control your desktop. Use a virtual environment and pinned dependencies, keep failsafe and approval mode enabled, avoid administrator execution, and do not use it around passwords, private chats, financial apps, admin consoles, or sensitive clipboard contents.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:93
Finding

Unpinned Third-Party Desktop Automation Dependencies

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:93 and duplicate instruction at SKILL.md:668
Vulnerability Type: Unpinned third-party dependency installation
Risk Level: Medium

Vulnerable Code Snippet:

bash
pip install pyautogui pillow opencv-python pygetwindow

Technical Analysis

The Skill directs the user or Agent to install four packages without exact version constraints, integrity hashes, a lockfile, or an explicitly trusted package index. Consequently, package resolution uses mutable package names and retrieves whichever compatible releases are available at installation time.

Python package installation may execute package-controlled installation or build logic. If a referenced package, one of its transitive dependencies, or the configured package repository is compromised, malicious code could execute with the privileges of the account running pip. The packages are especially security-sensitive because the Skill intends to use them for keyboard input, mouse control, screenshots, image recognition, and window management.

This finding does not establish that the named packages are currently malicious. The vulnerability is the absence of reproducible dependency resolution and integrity verification.

Attack Path

  1. An attacker compromises a referenced package, a transitive dependency, a future package release, or the package repository used by pip.
  2. A user or Agent follows the installation command in SKILL.md.
  3. pip resolves an unreviewed mutable release because no exact versions or hashes are specified.
  4. Package installation or build logic executes under the invoking user's security context.
  5. The compromised dependency obtains local code execution and may abuse the desktop-access capabilities expected by the Skill.

Impact Assessment

Successful exploitation could provide arbitrary code execution with the privileges of the account invoking pip. Potential scop ...[truncated 388 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin every direct dependency to an exact, reviewed version.
  2. Resolve and lock all transitive dependencies using a reproducible lockfile.
  3. Require cryptographic hashes, such as with pip install --require-hashes -r requirements.txt.
  4. Install only from an explicitly approved package index and prevent unintended fallback to additional indexes.
  5. Perform installation inside a dedicated virtual environment or sandbox using a non-administrative account.
  6. Review package provenance, release history, maintainers, and published artifacts before updating versions.
  7. Use automated dependency vulnerability and integrity scanning in the release process.
  8. Remove both unrestricted pip install examples from SKILL.md and replace them with installation instructions referencing the reviewed lockfile.
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger keywords include generic terms such as 'mouse', 'automation', 'advanced', 'control', and 'desktop', which are likely to match ordinary user requests unrelated to this specific skill. For a skill with powerful desktop-control and exec-enabled capabilities, overbroad triggering increases the chance of accidental invocation in sensitive contexts, leading to unintended system interaction or data exposure.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill description markets convenience and safety but does not prominently warn that it can capture screenshots, inspect windows, read clipboard contents, and simulate keyboard/mouse input on the host system. Because this is a high-privilege automation skill, the lack of clear privacy and system-impact warnings can cause users or agents to invoke it without understanding that it may access sensitive on-screen information or perform destructive actions.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The examples explicitly demonstrate screenshot capture, pixel inspection, image-based screen search, window enumeration, clipboard access, and automated form entry, all through simple natural-language-driven steps. In an agent context, these patterns lower the barrier to collecting sensitive on-screen data, copying confidential clipboard contents, or interacting with applications in ways that could leak secrets or alter system state without sufficiently strong safeguards.

Content

No source excerpt is available for this finding.