Security audit

Openclaw Unreal Skill

Security checks across malware telemetry and agentic risk

Overview

This skill is a disclosed Unreal Editor controller, but it exposes broad project-changing controls without built-in confirmation gates and has inconsistent safety guidance.

Install only for trusted local Unreal projects, keep disableModelInvocation enabled, and use source control/backups. Review every delete, save, import, console, input-simulation, screenshot, or log request before allowing the agent to run it.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands

Findings (6)

Description-Behavior Mismatch

Medium

Confidence: 94% confidence
Finding: The documentation explicitly recommends `disableModelInvocation=false`, encouraging autonomous AI-triggered use of powerful editor capabilities such as console execution, input simulation, actor deletion, asset import, screenshot capture, and log retrieval. This conflicts with the stated skill boundary that it should be used only on explicit Unreal Editor requests, and it increases the chance of unauthorized state changes or sensitive project data exposure without deliberate user approval.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The statement that the skill is 'automatically loaded' is ambiguous and does not clearly constrain when the skill should activate, which is risky for a skill that can create/delete actors, save levels, run console commands, and simulate input. In this context, vague activation behavior increases the chance of unintended tool invocation or social-engineering-driven misuse during ordinary Unreal-related conversations.

Missing User Warnings

Low

Confidence: 89% confidence
Finding: The README shows examples of state-changing Unreal Editor actions such as creating actors, starting play mode, taking screenshots, and moving objects, but it does not warn that these operations can modify project state or trigger runtime/editor-side effects. In the context of an agent skill that can control a local Unreal Editor, missing safety guidance increases the chance of accidental destructive actions or unintended project changes by users or downstream agents.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: The skill enumerates many destructive or state-changing operations—such as opening and saving levels, creating and deleting actors, modifying properties and transforms, adding/removing components, importing assets, running PIE, simulating input, and executing console commands—without prominent safety warnings about project corruption, unwanted edits, or disruptive runtime effects. In a tool that can directly control a local Unreal Editor, missing guardrails can lead to accidental destructive actions being normalized and executed without informed user consent.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: The documentation advertises screenshot capture and log retrieval without warning that these outputs may expose proprietary game content, local file paths, environment details, plugin names, errors, credentials accidentally logged by other systems, or other sensitive workstation/project information. Because the skill operates against a trusted local editor session, these capabilities are especially sensitive and can facilitate unintended disclosure if invoked automatically or without user awareness.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The tool exposes a broad command surface that includes destructive Unreal Editor actions such as actor deletion, level save/open, asset import, console execution, and input simulation, but the implementation performs no confirmation, allowlisting, or policy checks before queueing commands. In this skill context, the plugin is specifically designed to control a local Unreal Editor, so destructive actions are expected functionality, but the lack of user-facing safeguards makes accidental or unauthorized state-changing operations materially more dangerous.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.