Back to skill

Security audit

Claw Mouse

Security checks for vulnerabilities and agentic risk

Overview

The skill does what it says by automating an X11 desktop, but that gives an agent broad screen-capture and keyboard/mouse control with only minimal safety boundaries.

Review this carefully before installing. It is not evidence of malware, but it gives an agent the ability to view your screen, type, click, focus windows, and open URLs in a real X11 session. Use it only in a trusted, isolated desktop session where visible private data, credentials, and destructive applications are not exposed.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (12)

Env Variable Harvesting

High
Category
Data Exfiltration
Content
def run(ctx: Ctx, cmd, check=True, capture=False):
    env = os.environ.copy()
    env["DISPLAY"] = ctx.display
    env["XAUTHORITY"] = ctx.xauthority
    if capture:
Confidence
60% confidence
Finding
Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Lp3

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding
The skill exposes shell and environment-dependent desktop-control capabilities but does not declare any explicit tool scope or permission boundaries. In this context, the skill can capture screenshots and inject mouse/keyboard input into a live X11 session, so the absence of scoped permissions increases the risk of unintended credential capture, destructive GUI actions, or abuse by downstream agents.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Content
Ubuntu/Debian:
```bash
sudo apt-get update
sudo apt-get install -y xdotool scrot
```
Confidence
70% confidence
Finding
Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Content
Ubuntu/Debian:
```bash
sudo apt-get update
sudo apt-get install -y xdotool scrot
```
Confidence
70% confidence
Finding
Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

subprocess module call

Medium
Category
Dangerous Code Execution
Content
env["DISPLAY"] = ctx.display
    env["XAUTHORITY"] = ctx.xauthority
    if capture:
        return subprocess.run(cmd, env=env, check=check, text=True, capture_output=True)
    return subprocess.run(cmd, env=env, check=check)
Confidence
70% confidence
Finding
subprocess module calls execute external commands. Without careful input validation, this enables command injection.

subprocess module call

Medium
Category
Dangerous Code Execution
Content
env["XAUTHORITY"] = ctx.xauthority
    if capture:
        return subprocess.run(cmd, env=env, check=check, text=True, capture_output=True)
    return subprocess.run(cmd, env=env, check=check)


def require_bins(*bins):
Confidence
70% confidence
Finding
subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Content
raise SystemExit(
            "Missing required command(s): "
            + ", ".join(missing)
            + "\nInstall (Debian/Ubuntu): sudo apt-get install -y "
            + " ".join(missing)
        )
Confidence
70% confidence
Finding
Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The screenshot command captures and stores desktop images without any disclosure, consent check, or retention control. On a live desktop this can expose sensitive information visible on screen, including credentials, messages, tokens, or personal data, especially when used by an automated agent loop.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The click, type, and key functions provide direct GUI actuation with no confirmation, policy check, or safety interlock. In this context, that is powerful: an upstream agent can drive authentication dialogs, send messages, modify settings, or trigger destructive actions on the user's desktop.

Description-Behavior Mismatch

Medium
Confidence
93% confidence
Finding
The skill metadata describes screenshot and GUI input automation, but the code also includes an arbitrary URL/application launching primitive via xdg-open/gio/chromium-browser. That expands the capability beyond simple desktop control and can be abused to open attacker-controlled web content or trigger external handlers and applications unexpectedly.

Context-Inappropriate Capability

Medium
Confidence
88% confidence
Finding
Launching arbitrary URLs or applications is not clearly necessary from the stated purpose and materially increases the attack surface. In an agent-controlled desktop skill, this enables navigation to phishing pages, download prompts, or invocation of other registered application handlers without adequate constraint.

Missing User Warnings

Low
Confidence
79% confidence
Finding
Opening URLs through the system default handler can launch a browser or other external application without any user-facing warning. In an agentic desktop-control setting, even non-shell execution can still cause risky side effects by delegating to trusted local applications.

Static analysis

No suspicious patterns detected.